BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:KEYNOTE 7: The Future of Computing Will Be non-von Neumann DTSTART;VALUE=DATE-TIME:20191204T160000Z DTEND;VALUE=DATE-TIME:20191204T164500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1091@events.chpc.ac.za DESCRIPTION:Speakers: Thomas Sterling (Indiana University)\nSeven decades of HPC has been empowered by the abstraction of “the von Neumann Archite cture” and its many derivatives\, driven to a significant degree by Moor e’s Law and the exponential growth of device density and concomitant clo ck rates yielding a performance gain over that period of more than ten tri llion for floating point computation. But a perfect storm of recent techno logy trends has terminated this unprecedented expansion and challenges the future opportunities beyond exascale computing. But even as the end of co nventional processing practices is flat-lining\, a new world of non von Ne umann execution models and architectures is emerging igniting a revolution for the next generations of computing systems orders of magnitude greater performance than is currently achieved. Even more important is that the r eality of HPC users is that approximately 90% of the Top-500 machines meas ured with the HPL benchmark demonstrate only about 1% of the performance o f the fastest machines. Thus we are much further away from exascale than i s generally assumed and therefore much greater gains are required to truly bring the major base of HPC users into the exascale era. New non von Neum ann architectures and models\, such as Quantum Computing\, Neomorphic Comp uting\, and Continuum Computing (this last presented at CHPC18) are offeri ng important possibilities for the future. Earlier non von Neumann techniq ues previously explored in past decades\, such as static and dynamic dataf low\, cellular automata\, systolic arrays\, and special purpose designs ma y also serve as starting points for new classes of useful computing method s even as Moore’s Law recedes. Finally\, advanced technologies beyond co nventional semiconductors such as cryogenic such as single flux quantum lo gic provides yet another dimension of potential post exascale strategies. This Keynote Address will convey a fast paced odyssey through the near fut ure opportunities of HPC non von Neumann based computers. Questions will b e encouraged by participants throughout the presentation as well as the Q& A session at its conclusion.\n\nhttps://events.chpc.ac.za/event/47/contrib utions/1091/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1091/ END:VEVENT BEGIN:VEVENT SUMMARY:BRICS programme on “Cybersecurity: Software and Data Security” DTSTART;VALUE=DATE-TIME:20191203T092500Z DTEND;VALUE=DATE-TIME:20191203T095000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1084@events.chpc.ac.za DESCRIPTION:Speakers: Joey Jansen van Vuuren (Tshwane University of Techno logy)\nhe BRICS Network University (BRICS NU) is a network of 60 universit ies\, 12 each from the five BRICS countries. The BRICS NU is aimed at deve loping partnerships and exchange programmes in six thematic areas (ITGs) d etermined by the BRICS Ministries responsible for education. This projec t forms part of the University Capacity Development Programme (UCDP)\, an umbrella imitative developed and implemented by the South African Departme nt of Higher Education and Training to build capacity in South African uni versities in three key areas: student success\, staff development and prog ramme/curriculum development. The BRICS NU ITG on Computer Science and Inf ormation Security (CSIS) representatives from the BRICS countries\, agreed that CSIS ITG will focus on developing a BRICS Masters programme on ìCyb ersecurity: Software and Data Security. The presentation will include the proposed content\, development and implementation plan of the BRICS Master s Programme.\n\nhttps://events.chpc.ac.za/event/47/contributions/1084/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1084/ END:VEVENT BEGIN:VEVENT SUMMARY:SA NREN Talk 3 DTSTART;VALUE=DATE-TIME:20191203T095000Z DTEND;VALUE=DATE-TIME:20191203T101500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1085@events.chpc.ac.za DESCRIPTION:Speakers: Attlee Munyaradzi Gamundani ()\nhttps://events.chpc. ac.za/event/47/contributions/1085/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1085/ END:VEVENT BEGIN:VEVENT SUMMARY:SANReN Data Transfer Pilot DTSTART;VALUE=DATE-TIME:20191203T101500Z DTEND;VALUE=DATE-TIME:20191203T102500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1083@events.chpc.ac.za DESCRIPTION:Speakers: Kasandra Pillay (SANReN)\nMoving masses of data is a challenge. It can be slow and frustrating to transfer vast quantities of data from the many places it can be stored or generated over general-purpo se computer networks.\n\nWhen scientists attempt to run data intensive app lications over campus networks\, it often results in slow transfers - in m any cases poor enough that the science mission is significantly impacted. In the worst case\, this means either not getting the data\, getting it to o late or resorting to alternative inefficient measures such as shipping d isks around.\n\nSANReN would like to provide an update on the Data Transfe r Pilot service that is available for the South African Research and Educa tion community\, to assist them to move their data locally and internation ally. This session will also provide Data Transfer updates within the Nati onal Integrated Cyber Infrastructure System (NICIS).\n\nhttps://events.chp c.ac.za/event/47/contributions/1083/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1083/ END:VEVENT BEGIN:VEVENT SUMMARY:BIOPERIANT12\, a regional high-resolution model configuration towa rds developing the South African VrESM Earth System Model DTSTART;VALUE=DATE-TIME:20191203T121000Z DTEND;VALUE=DATE-TIME:20191203T123000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1081@events.chpc.ac.za DESCRIPTION:Speakers: Nicolette Chang (Ocean Systems and Climate\, CSIR)\n In understanding and predicting a changing global climate system\, the rep resentation of ocean-biogeochemistry processes in the Southern Ocean is pa rticularly important because of the key role it plays in global carbon-cl imate feedbacks. To date\, Earth System Models (ESMs) do not adequately r esolve important ocean dynamics (e.g.\, mesoscale processes)\, features th at are critical in Southern Ocean heat and CO2 fluxes and storage. Therefo re high resolution ocean biogeochemical models provide essential constrain ts to the medium resolution (100km) global ESMs.\n\nThe South African ESM\ , VrESM\, comprises of globally coupled atmosphere\, ocean\, ice\, land-su rface\, atmospheric-chemistry\, and ocean-biogeochemistry models. Building and running the ESM is therefore a huge task: both scientifically and com putationally. Several numerical models\, each discretized on a global grid need to be integrated in space and time\, while additionally passing info rmation to each other. As part of a multi-institution and multi-year goal of building South Africa’s first Earth System Model\, which will be run at the CHPC\, we have been developing the ocean-biogeochemistry component of the VrESM (PISCES-SOCCO). BIOPERIANT12 is a critical platform in this d evelopment. \n\nWe present the NEMO v3.4 regional model configuration BIOP ERIANT12\, our most computationally-challenging model to date and run on C HPC’s Lengau cluster. BIOPERIANT12 simulates ocean\, ice\, biogeochemist ry of the circumpolar Southern Ocean (south of 30°S) from 1989 to 2009\, prescribed by ERA-interim atmospheric forcing. BIOPERIANT12 is high resolu tion at a mesoscale-resolving 8 km in the horizontal and in the vertical: ranges from 6 m resolution at the surface to 250 m at the ocean bottom ov er 46 vertical levels. \n\nIn addition to the technical aspect of developi ng the PISCES-SOCCO source code for VrESM\, we have to configure VrESM fo r an improved representation of the Southern Ocean. BIOPERIANT12\, thus se rves in multiple ways: (1) as a comparison for ocean biogeochemistry in th e ESM\, (2) as a large test case for ocean-biogeochemical evaluation metri cs for the ESM\, (3) as an experimental platform for understanding process es which influence atmosphere-ocean carbon exchange in the Southern Ocean\ , which additionally helps improve the ESM. We discuss PISCES-SOCCO develo pment progress as well as the building and evaluation of BIOPERIANT12.\n\n https://events.chpc.ac.za/event/47/contributions/1081/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1081/ END:VEVENT BEGIN:VEVENT SUMMARY:Interaction between S. mansoni Universal stress G4LZI3 protein and selected polyphenols: a bioinformatics investigation. DTSTART;VALUE=DATE-TIME:20191202T123000Z DTEND;VALUE=DATE-TIME:20191202T125000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1080@events.chpc.ac.za DESCRIPTION:Speakers: Abidemi Paul Kappo (University of Zululand\, KwaDlan gezwa 3886\, South Africa)\nFor decades\, Praziquantel has been the undisp uted drug of choice against schistosomiasis\; a disease that affects more than 200 million people in over 78 countries and responsible for over 280 000 lives lost per annum\, predominantly in sub-Saharan Africa. Rising con cerns have been raised due to the unknown mechanism of action of the drug and unavoidable reports of the emergence of drug resistant strains. Moreov er\, current apprehension has been reinforced by the total dependence on a single drug for treatment. Therefore\, the search for novel and effective anti-schistosomal drugs become imperative. This study made use of bioinfo rmatics tools to determine the binding properties of a selective range of polyphenols docked onto the Universal stress G4LZI3 protein\, a recently i dentified ‘lead’ molecule in the design of alternative treatment drug against schistosomiasis. Schistosomes have over several years\, evolved me chanisms that include the presence of USPs\, to counter biotic and abiotic stress. Up-regulation of the G4LZI3 protein throughout the multifaceted d evelopmental cycle of the schistosome worm sparks interest in this protein \, whose function is currently unknown. Ten polyphenols were docked onto t he G4LZI3 protein\; the best five complexes were selected for post-molecul ar dynamics analyses and binding free energy calculations. The strongest b inding interactions were observed between the G4LZI3 protein with curcumin and catechin respectively. The major interacting residues conserved in al l the complexes provides basis for further structure-based drug design of new compounds\, with enhanced inhibitory potency and toxicity against G4LZ I3. This study suggests an alternative approach for the development of ant i-schistosomal drugs using natural compounds.\n\nhttps://events.chpc.ac.za /event/47/contributions/1080/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1080/ END:VEVENT BEGIN:VEVENT SUMMARY:High Performance Computing for Medical Interventional Planning App lications DTSTART;VALUE=DATE-TIME:20191202T090000Z DTEND;VALUE=DATE-TIME:20191202T092000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1078@events.chpc.ac.za DESCRIPTION:Speakers: Malebogo Ngoepe (UCT)\nPersonalised solutions for he althcare are increasingly recognised as an important approach for treating a variety of conditions that have different outcomes\, based on the patie nt. In the field of computational mechanics\, different virtual pipelines have been developed in an effort to improve interventional planning and lo ng-term patient outcomes. One of the major challenges to realising patient -specific treatment tailoring is a mismatch of timeframes. Clinical diagno ses and treatments need to be carried out in as short a timeframe as possi ble\, while traditional CFD codes tend to run over longer time periods.\n\ nThe use of high performance computing platforms has been beneficial in th e development of interventional planning pipelines for cerebral aneurysm t hrombosis and congenital heart disease. For aneurysms\, it is important t o determine what type of clot will form in the aneurysm sac\, based on the treatment modality selected. In the case of congenital heart disease\, tr eatments which are selected need to be optimised to ensure that solutions will remain suitable as the child grows to adulthood. This talk will explo re the challenges encountered in developing these two pipelines for clinic al use.\n\nhttps://events.chpc.ac.za/event/47/contributions/1078/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1078/ END:VEVENT BEGIN:VEVENT SUMMARY:Predicting Thermodynamic Properties of Ionic Liquids—from Molecu lar Simulation to Machine Learning DTSTART;VALUE=DATE-TIME:20191202T121000Z DTEND;VALUE=DATE-TIME:20191202T123000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1049@events.chpc.ac.za DESCRIPTION:Speakers: Gerhard Venter (University of Cape Town)\nRoom Tempe rature Ionic Liquids (RTILs)\, which are broadly defined as salts that are liquid below 100 °C\, offer an alternative to typical organic solvents. Favourable properties\, such as high thermal stability\, large melting ran ge\, low vapour pressure (and consequently low flammability) and miscibili ty with both polar and nonpolar compounds\, provide the motivation for the se systems to become next-generation solvents for synthesis\, catalysis an d separation technologies. Since all ILs consist of both a cation and anio n\, interchange between existing and novel ions can deliver a vast set of new solvent systems\, much quicker than derivatisation of regular\, molecu lar systems. Consequently\, the millions of potential systems for explorat ion make property prediction of novel ILs an exciting research opportunity . Coupled with the global demand to reduce waste and develop processes wit hin the principles of sustainable and green chemistry\, there is increasin g interest in moving away from conventional solvent systems.\nTo speed up rational design methodologies and avoid costly synthetic routes\, cost-eff ective alternatives to experimental screening of potential new ILs for fav ourable physical and mechanical properties are needed. The estimation of t hermodynamic properties is an essential component of this task and the app lication of high-performance computing is central to achieving this. Estim ation methods can come in various forms but can be divided into three broa d categories: (1) molecular/atomistic simulation and computation\, (2) emp irical modelling (e.g. equations of state) or (3) quantitative structure p roperty relationships\, driven by machine learning approaches. The first r equires a classical or quantum mechanical description of molecules and a t heoretical framework (such as statistical mechanics) to derive macroscopic properties from these\; the second is immersed in classical thermodynamic s and exploits the relationships between state variables\; the last finds patterns in data and can build linear or nonlinear functions of user-defin ed features to express physical properties without the need for direct\, p rior knowledge of these. Each comes with its own strengths and weaknesses in terms of resources needed (empirical information and computational cost )\, domain of applicability\, accuracy and ease-of-use.\nIn this presentat ion\, the prediction power of molecular simulation and machine learning me thods as applied to ILs is put to the test. The constant pressure heat cap acity is chosen as the target—this property expresses the response of a system’s energy to a temperature change and can be both quantified theor etically and measured experimentally with reasonable ease. A test set of f ive structurally diverse ionic liquids have been picked and their temperat ure dependent heat capacities calculated using classical molecular dynamic s (MD) simulations with various force field implementations as well as a s election of machine learning algorithms. In addition to discussing the acc uracy\, the strengths and weaknesses of the difference approaches are also compared.\nMD simulations\, of systems consisting of 10k to 20k atoms\, w ere run using the AMBER code\, which has been implemented on both GPU and CPU architectures. The former is highly efficient and speedups in excess o f 20x can be obtained (> 500 ns/day). The CPU code runs in parallel using an MPI implementation that scales well up to 64 processors (system size de pendent). The Scikit-learn Python framework was used for machine learning model development with the Keras API and TensorFlow as backend for the art ificial neural network models. In addition to built-in support for multit hreading\, where applicable\, embarrassingly parallel steps in model train ing and validation were optimized using the mpi4py module.\n\nhttps://even ts.chpc.ac.za/event/47/contributions/1049/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1049/ END:VEVENT BEGIN:VEVENT SUMMARY:ML approaches for HPC DTSTART;VALUE=DATE-TIME:20191203T093000Z DTEND;VALUE=DATE-TIME:20191203T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1043@events.chpc.ac.za DESCRIPTION:Speakers: Peter Braam (University of Oxford)\nMachine Learning methodologies and tools have delivered new approaches to scientific compu ting ranging from new approximation methods to solve differential equation s to leveraging advantages of ML hardware over traditional HPC hardware. I t is not unlikely that such approaches will be helpful to computational pr oblems that have seen little progress for decades. We will discuss a few examples\, and discuss key themes in carrying this forward.\n\nhttps://eve nts.chpc.ac.za/event/47/contributions/1043/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1043/ END:VEVENT BEGIN:VEVENT SUMMARY:KEYNOTE 6: Eliminating Weapons of Math Destruction: Next-Generati on Arithmetic DTSTART;VALUE=DATE-TIME:20191204T074500Z DTEND;VALUE=DATE-TIME:20191204T083000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1069@events.chpc.ac.za DESCRIPTION:Speakers: John Gustafson (National University of Singapore)\nU sers of floating-point arithmetic (floats) have long experienced the disco nnect between mathematically correct answers and what a computer provides. Choices made in the 1986 IEEE 754 Standard for floats lead to irreproduci ble results that destroy the confidence we experience\, say\, when working with integers. After 33 years\, language support for mandated internal fl ags (rounding\, overflow\, etc.) remains nil\, so float hazards are almost invisible. The Standard does not require correct or consistent rounding o f transcendental functions\, so bitwise portability of float-based program s is nonexistent.\n \nThe emerging posit standard is a fresh approach to c omputing with real numbers that is fast\, bitwise-reproducible\, and capab le of preserving mathematical properties like the associative and distribu tive laws of algebra without sacrificing performance. Complete hardware-so ftware stacks supporting this new kind of arithmetic are beginning to appe ar\, so we now have the hope of eliminating IEEE 754 "weapons of math dest ruction" with something much closer to the logical behavior we expect from computers.\n\nhttps://events.chpc.ac.za/event/47/contributions/1069/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1069/ END:VEVENT BEGIN:VEVENT SUMMARY:Benchmarking off-the-shelf radio astronomy software on Lengau DTSTART;VALUE=DATE-TIME:20191203T115000Z DTEND;VALUE=DATE-TIME:20191203T121000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1057@events.chpc.ac.za DESCRIPTION:Speakers: Sean February (SARAO)\nMeerKAT\, one of the world’ s most powerful radio telescopes in operation today\, is producing science -ready data into the now public-facing archive at a steady rate. In this t alk\, we will assess the performance of off-the-shelf tools for post-proce ssing and imaging MeerKAT data on Lengau.\n\nhttps://events.chpc.ac.za/eve nt/47/contributions/1057/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1057/ END:VEVENT BEGIN:VEVENT SUMMARY:Opening Session DTSTART;VALUE=DATE-TIME:20191202T070000Z DTEND;VALUE=DATE-TIME:20191202T074500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1071@events.chpc.ac.za DESCRIPTION:https://events.chpc.ac.za/event/47/contributions/1071/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1071/ END:VEVENT BEGIN:VEVENT SUMMARY:Prize Giving Ceremony DTSTART;VALUE=DATE-TIME:20191204T164500Z DTEND;VALUE=DATE-TIME:20191204T173000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1075@events.chpc.ac.za DESCRIPTION:https://events.chpc.ac.za/event/47/contributions/1075/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1075/ END:VEVENT BEGIN:VEVENT SUMMARY:The adoption of Deep Learning in Weather Forecast DTSTART;VALUE=DATE-TIME:20191203T123000Z DTEND;VALUE=DATE-TIME:20191203T125000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1058@events.chpc.ac.za DESCRIPTION:Speakers: Yania Molina Souto (LNCC)\nTechnological development and the internet of things (IoT) increased the data assimilation sources in meteorology through satellites\, sensors\, weather stations\, solar pan els\, cell phones\, traffic lights\, to name a few whose number grows dail y. This begins to build an ideal scenario for Artificial Intelligence wher e the demand for data is high. This combined information generates spatio- temporal maps of temperature\, rainfall\, air movement\, etc.\, with high precision in regions with larger data sources. Applying AI techniques in c onjunction with a physical understanding of the environment can substantia lly improve prediction skill for multiple types of high-impact weather eve nts.\n\nIn 2017\, the American Meteorological Society (AMS) published a pa per [1] with a broad summary indicating how modern AI techniques are helpi ng to improve insights and make decisions in weather prediction. Among the most used techniques are Support Vector Machines (SVM)\, regression trees \, k-means for radar image segmentation and traditional neural networks (A NN). These techniques lack temporal and spatio-temporal analysis\, typical of meteorological phenomena. Today\, most of the temporal analysis done o n meteorological data is through statistical algorithms\, such as autoregr essive methods. However\, the AMS recognizes that the novel Deep Learning techniques could soon be the cause of new improvements and says\, “In th e future\, convolutional neural networks operating in a deep learning fram ework may reduce the need for feature engineering even further” .\n\nCom plementarily\, recurrent neural networks\, designed for analysis of natura l language processing\, are known by their results in numerical problems l ike temperature prediction\, among others. In order to improve the tempora ry forecasts of spatio-temporal phenomena\, such as rain and temperature\, hybrid architectures have emerged that build the temporary forecast codin g the spatial pattern of the neighborhood.\n\nIn [7]\, Shi et al. formulat e precipitation nowcasting as a spatio-temporal sequence forecasting probl em in which both the input and the prediction target are spatio-temporal s equences. They extend the fully connected LSTM (FC-LSTM) to have convoluti onal structures in both the input-to-state and state-to-state transitions\ , they propose the convolutional LSTM (ConvLSTM) and use it to build an en d-to-end trainable model for the precipitation nowcasting problem. Experim ents show that the ConvLSTM network captures spatio-temporal correlations better and consistently outperforms FC-LSTM and the state-of-the-art opera tional ROVER algorithm for precipitation nowcasting.\n\nIn [8]\, Souto et al. use a ConvLSTM architecture as a spatio-temporal ensemble approach. Th e channels in the convolution operator are used to input different physica l weather models. In this way the ConvLSTM encodes the spatial information which are subsequently learned by the recurrent structures of the network . The results show that ConvLSTM achieves superior improvements when compa red to traditionally used ensemble techniq ues such as BMA [9].\n\nAs a ma tter of fact\, there are a plethora of opportunities to be investigated ex tending the initial results we have achieved in adopting Deep Neural netwo rks to weather prediction. Linear and causal convolution operators (the la tter also known as temporal convolution)\, for instance\, have resulted in deep networks architectures that use convolutions to encode and decode ti me and space with greater precision. Raissi and colleagues [10] investigat e the integration of physical laws described as a set of partial different ial equations to the training process. By means of such integration\, the training process is bound to obey the physical laws\, an approach that has been dubbed as model teaching . Another area of interest is multimodal ma chine learning (MML) [11]. In MML\, data from different representations ar e complementarily used in building models\, including: images\; textual da ta\; quantitative dataetc... This can be extremely interesting in weather forecast as more data is captured from satellite images and sensors data t o weather bulletins and predictive physical models.\n\nREFERENCES\n1. Amy McGovern\, Kimberly L. Elmore\, David John Gagne II\, Sue Ellen Haupt\, Ch ristopher D. Karstens\, Ryan Lagerquist\, Travis Smith\, and John K. Willi ams. Using Artificial Intelligence to Improve Real-Time Decision-Making fo r High-Impact Weather. AMS. 2017\n2. Qi Fu\, Dan Niu\, Zengliang Zang\, Ju nhao Huang\, and Li Diao. Multi-Stations’ Weather Prediction Based on Hy brid Model Using 1D CNN and Bi-LSTM. Chinese Control Conference (CCC)\, 37 71-3775. 2019\n3. Anthony Wimmers\, Christopher VThomas Bolton and Laure Z anna . Applications of Deep Learning to Ocean Data Inference and Subgrid P arameterization. Journal of Advances in Modeling Earth Systems 11:1\, 376- 399. 2019\n4. Anthony Wimmers and Christopher Velden \, and Joshua H. Coss uth . Using Deep Learning to Estimate Tropical Cyclone Intensity from Sate llite Passive Microwave Imagery. Monthly Weather Review 147:6\, 2261-2282. 2019\n5. S. Scher. Toward Data-Driven Weather and Climate Forecasting: Ap proximating a Simple General Circulation Model With Deep Learning. Geophys ical Research Letters 45:22\, 12\,616-12\,622. 2018\n6. Brian S. Freeman\, Graham Taylor\, Bahram Gharabaghi and Jesse Thé. Forecasting air quality time series using deep learning. Journal of the Air & Waste Management As sociation\, Volume 08\, Issue 8 \, Pages 866-886\, 2018\n7. Xingjian Shi\, Zhourong Chen\, and Hao Wang. Convolutional LSTM Network : A Machine Lear ning Approach for Precipitation Nowcasting. arXiv\, pages 1–11\, 2015.\n 8. Yania Molina Souto\, Fábio Porto \, Ana Maria de Carvalho Moura \, Edu ardo Bezerra : A Spatiotemporal Ensemble Approach to Rainfall Forecasting. IJCNN \, pages 1-8. 2018\n9. Raftery\, A. E.\, Gneiting\, T.\, Balabdaoui \, F.\, and Polakowski\, M.: Using Bayesian model averaging to calibrate f orecast ensembles\, Mon. Weather Rev.\, 133\, 1155–1174\, 2005.\n10.Rais si\, M.\, Predikaris\, P.\, Karniadakis\, G.E.\, Physics-informed neural n etwork: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computatio nal Physics\, 378\, pp.686-707\, 2019.\n11.Tadas Baltrusaitis\, Chaitanya Ahuja\, and Louis-Philippe Morency\, MultiModal Machine Learning: A Survey and Taxonomy\, https://arxiv.org/pdf/1705.09406.pdf\n\nhttps://events.chp c.ac.za/event/47/contributions/1058/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1058/ END:VEVENT BEGIN:VEVENT SUMMARY:The research benefits of improving equity\, diversity and inclusio n DTSTART;VALUE=DATE-TIME:20191203T113000Z DTEND;VALUE=DATE-TIME:20191203T115000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1056@events.chpc.ac.za DESCRIPTION:Speakers: Toni Collis (WHPC & Collis-Holmes Innovations)\nEqui ty\, diversity and inclusion seem to be the magic words of the 21st Centur y. We’ve all been told they are important\, but for many\, it is nice-to -have but difficult-to-implement. As there is an increasing shortage of hi ghly skilled technology specialists and an increasingly large array of app lications\, it has never been more important to implement workable equity\ , diversity and inclusion practices to ensure that we can attract and reta in talent \n \nThis session will discuss the benefits of diversity and wha t we all need to be doing to ensure our community benefits. Discussions wi ll include the work being carried out by Women in High Performance Computi ng to diversify the international supercomputing community and what the Af rican HPC community can learn from this.\n\nhttps://events.chpc.ac.za/even t/47/contributions/1056/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1056/ END:VEVENT BEGIN:VEVENT SUMMARY:MeerKAT archive and data access DTSTART;VALUE=DATE-TIME:20191202T093000Z DTEND;VALUE=DATE-TIME:20191202T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-959@events.chpc.ac.za DESCRIPTION:Speakers: Martin Slabber (SARAO)\nThe MeerKAT archive was made accessible from the internet earlier this year.\nThis allow researchers f rom across the world to pull data from the MeerKAT archive.\nIn this talk I'll describe the MeerKAT storage system that consists out of several peta bytes of observations backed by a Ceph distributed storage system. The use of Ceph at SARAO and the infrastructure around the archive will be descri bed. There after the data access methods from the internet and from partne r institutes like the CHPC will be presented.\n\nhttps://events.chpc.ac.za /event/47/contributions/959/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/959/ END:VEVENT BEGIN:VEVENT SUMMARY:Computational Fluid Dynamics: from solar thermal receivers and col lectors to sub-antarctic island wind simulation DTSTART;VALUE=DATE-TIME:20191202T115000Z DTEND;VALUE=DATE-TIME:20191202T121000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1048@events.chpc.ac.za DESCRIPTION:Speakers: Ken Craig (University of Pretoria)\nThe parallel clu ster at the CHPC is used to provide speed-up on three fronts in the Comput ational Fluid Dynamics (CFD) simulations discussed. The first is for optim ization of the parameterized geometry of a swirling jet impingement solar thermal receiver where many runs of fairly large Large Eddy Simulation CFD models are required. The second problem that benefits from the massively parallel approach\, is that of a transient simulation of the atmospheric b oundary layer turbulent flow field around a heliostat with a very small ti me step\, requiring many time steps for a meaningful time series. This sim ulation is required to determine peak loads and perform the fluid-structur e interaction of such a solar collector. The last type is for large models of wind flow over Marion island containing close to hundred million compu tational cells. These models are used to predict wind patterns that affect plant and bird life\, especially as influenced by continued climate chang e.\n\nhttps://events.chpc.ac.za/event/47/contributions/1048/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1048/ END:VEVENT BEGIN:VEVENT SUMMARY:Design of selective reagents and functional materials: A theoretic al and experimental approach DTSTART;VALUE=DATE-TIME:20191203T090000Z DTEND;VALUE=DATE-TIME:20191203T092000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1052@events.chpc.ac.za DESCRIPTION:Speakers: Zenixole Tshentu (Nelson Mandela University)\nThe wo rk illustrates the integration of basic and applied chemistry in the devel opment and application of selective materials for application in desulfuri zation and denitrogenation of fuel as well as in separation of precious me tals. The fuel chemistry study is important from the point of view of the need to drive towards a zero sulfur fuel as mandated by environmental prot ection agencies in many countries around the world.1 Challenges exist with the current hydrodesulfurization and hydrodenitrogenation processes that are being applied in refineries as they fail to achieve the requisite fuel standards. The second application of functional materials is in separatio n of important metals. The demand for precious metals is driven by their i mportant applications\, and the development of better separating reagents/ materials has become important given that the quality of ores is decreasin g\, and better recovery rates of the metals from secondary sources (such a s electronic boards and catalytic converters) will be required in future. This necessitates improvement of the current chemistry in order to process the new feeds. \n\nExperimental and theoretical studies were carried out during the development of the functional chemistry for recognition of targ et metals and organic compounds. The selective chemistry towards fuel con taminants such as organosulfur and organonitrogen compounds has been devel oped\, and the results are promising as the best material (polymenzimidazo le nanofibers) achieve sulfur removal of less than 2 ppm.2 A process invol ving conversion of sulfur compounds to organosulfones compounds3 has been developed followed by removal of the polar sulfones using selective materi als.2 The approach for materials development for metal ions\, such as plat inum group metals (PGMs)\, follows the development of reagents that are sp ecific for metal ion chlorido complexes of interest. The innovation of the aforesaid reagents undoubtedly requires a design strategy that considers both the electronic and stereochemical requirements of the target anion. T hrough a combination of molecular modelling techniques and experimental te chniques\, we have been able to derive factors that lead to successful sep arations. Cations as anion receptors specific for [IrCl6]2- and [PtCl6]2 ⁻ will be presented as well as selective chemistry for orgnosulfur and o rganonitrogen compounds in fuel. Binding energies and other thermodynamic parameters have been calculated in silico to explain the chemistry involve d.\n\nReferences: \n1. Barbara\, P.\; Rufino\, M.N.\; Campos-Martin\, J.M. \; Fierro\, J.L.G.\; Catal. Sci. Technol.\, 2011\, 1\, 23-42. \n2. Ogunla ja\, A.S.\; du Sautoy\, C.\; Torto\, N.\; Tshentu\, Z.R.\; Talanta\, 2014\ , 126\, 61–72.\n3. Ogunlaja\, A.S.\; Chidawanyika\, W.\; Antunes\, E.\; Fernandes\, M.A.\; Nyokong\, T.\; Torto\, N.\; Tshentu\, Z.R.\; Dalton Tra ns\, 2012\, 41\, 13908-13918.\n\nhttps://events.chpc.ac.za/event/47/contri butions/1052/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1052/ END:VEVENT BEGIN:VEVENT SUMMARY:SAVIME - Simulation Analysis and Visualization in-Memory DTSTART;VALUE=DATE-TIME:20191203T092000Z DTEND;VALUE=DATE-TIME:20191203T094000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1053@events.chpc.ac.za DESCRIPTION:Speakers: Fabio Porto (LNCC)\nThe increasing computational pow er of HPC systems fosters the\ndevelopment of complex numerical simulation s of phenomena in different\ndomains\, such as medicine [1]\, Oil & Gas [2 ] and many other fields [3\,4\,5].\nIn such applications\, a huge amount o f data in the form of multidimensional\narrays is produced and need to be analyzed and visualized enabling\nresearchers to gain insights about the p henomena being studied.\n\nScientists also generate huge multidimensional arrays through\nenvironmental observations\, measurements of physical cond itions and\nother types of sensors. For instance\, satellite data for Eart h's weather\,\noceans\, atmosphere and land [6] are kept in the form of mu ltidimensional\narrays in scientific file formats. Data collected by senso rs in physics\nexperiments\, such as the ones conducted in the photon stud ies by SLAC\nNational Accelerator Laboratory [7]\, are also represented an d processed in\nthe form of multidimensional arrays.\n\nMachine learning i s another context in which multidimensional arrays are\npresent. They are the basic input format for the heavily optimized linear\nalgebra algorithm s implemented in deep learning frameworks\, such as:\nTensorFlow\, Keras a nd Torch. Deep Learning algorithms were able to\nachieve superhuman perfor mance for image recognition problems in the\npast few years [8]\, and they are among the most promising alternative for\ntackling difficult problems in Natural Language Processing\, Image and\nVideo Recognition\, Medical I mage Analysis\, Recommendation Systems\nand many others. Thus\, managing t hese large arrays in the context of deep\nlearning is a very important tas k.\n\nThe traditional approach for managing data in multidimensional array s in\nscientific experiments is to store them using file formats\, such as netCDF\nand HDF5. The use of file formats\, and not a database management \nsystems (DBMS)\, in storing scientific data has been the traditional cho ice\ndue to the fact that DBMSs are considered inadequate for scientific d ata\nmanagement. Even specialized scientific data management systems\, suc h\nas SciDB [10]\, are not very well accepted for a myriad of reasons list ed in\n[11\, such as]:\n\n● the impedance mismatch problem [12\,13]\, th at makes the process of\ningesting data into a DBMS very slow.\n● the in ability to directly access data from visualization tools like\nParaview Ca talyst [14] and indexing facilities like FastQuery [13].\n● the Inabilit y to directly access data from custom code\, which is\nnecessary for domai n specific optimized data analysis.\n\nHowever\, by completely dismissing DBMSs\, some nice features also\nbecome unavailable. Including the access for out-the-box parallel\ndeclarative data processing with the usage of qu ery languages and query\noptimization\, and management of dense and sparse matrices. In this talk\,\nwe will present SAVIME\, a Database Management System for Simulation\nAnalysis and Visualization in-Memory. SAVIME implem ents a\nmulti-dimensional array data model and a functional query language . The\nsystem is extensible to support data analytics requirements of nume rical\nsimulation applications.\n\nReferences\n[1] Pablo J. Blanco\, M. R. Pivello\, S. A. Urquiza\, and Raúl A. Feijóo. 2009.\nOn the potentialit ies of 3D-1D coupled models in hemodynamics\nsimulations.Journal of Biomec hanics 42\, 7 (2009)\, 919–930.\n[2] Cong Tan et al. 2017. CFD Analysis of Gas Diffusion and Ventilation\nProtection in Municipal Pipe Tunnel. In Proceedings of the 2017\nInternational Conference on Data Mining\, Communi cations and Information\nTechnology (DMCIT ’17). ACM\, New York\, NY\, U SA\, Article 28\, 6 pages.\nHttps: //doi.org/10.1145/3089871.3089904\n[3] Igor Adamovich et al. 2015. Kinetic mechanism of molecular energy\ntransfe r and chemical reactions in low-temperature air-fuel plasmas.\nPhilosophic al transactions. Series A\, Mathematical\, physical\, and\nengineering sci ences 373 (08 2015).\nhttps://doi.org/10.1098/rsta.2014.0336\n[4] Mollona\ , Edoardo\, Computer simulation in social sciences\, Journal of\nManagemen t & Governance\, May\, 2008\, N(2) V(12).\n[5] Mitsuo Yokokawa\, Ken’ich i Itakura\, Atsuya Uno\, Takashi Ishihara\, and\nYukio Kaneda. 2002. 16.4T flopss Direct Numerical Simulation of\nTurbulence by a Fourier Spectral Me thod on the Earth Simulator. In\nProceedings of the 2002 ACM/IEEE Conferen ce on Supercomputing (SC\n’02). IEEE Computer Society Press\, Los Alamit os\, CA\, USA\, 1–17.\nhttp://dl.acm.org/citation.cfm?id=762761.762808 . \n[6] R. Ullman and M. Denning. 2012. HDF5 for NPP sensor and\nenvironment al data records. In 2012 IEEE International Geoscience and\nRemote Sensing Symposium. 1100–110\n[7] Jack Becla\, Daniel Wang\, and Kian-Tat lim. 2 011. Using SciDB to\nSupport Photon Science Data Analysis. (01 2011).\n[8] Dan Ciresan\, Alessandro Giusti\, Luca M. Gambardella\, and Jürgen\nSch midhuber. 2012. Deep Neural Networks Segment Neuronal\nMembranes in Electr on Microscopy Images. In Advances in Neural\nInformation Processing System s 25 \, F. Pereira\, C. J. C. Burges\, L. Bottou\,\nand K. Q. Weinberger ( Eds.). Curran Associates\, Inc.\, 2843–2851.\n[9] Z Zhao. 2014. Automati c library tracking database at NERSC. (2014).\nhttps://www.nersc.gov/asset s/altdatNERSC.pdf\, J. ACM\, Vol. 37\, No. 4\,\nArticle 111. Publication d ate: August 2019.\n[10] Paradigm4. 2019. SciDB. http://www.paradigm4.com/ [Online\;\naccessed 01-sep-2019].\n[11] Haoyuan Xing\, Sofoklis Floratos\, Spyros Blanas\, Suren Byna\, Prabhat\,\nKesheng Wu\, and Paul Brown. 2017 . ArrayBridge: Interweaving declarative\narray processing with high-perfor mance computing. arXiv e-prints \, Article\narXiv:1702.08327 (Feb 2017).\n [12] Spyros Blanas\, Kesheng Wu\, Surendra Byna\, Bin Dong\, and Arie\nSho shani. 2014. Parallel Data Analysis Directly on Scientific File Formats.\n In Proceedings of the 2014 ACM SIGMOD International Conference on\nManagem ent of Data (SIGMOD ’14) . ACM\, New York\, NY\, USA\, 385–396.\n[13] Luke Gosink\, John Shalf\, Kurt Stockinger\, Kesheng Wu\, and Wes\nBethel. 2006. HDF5-FastQuery: Accelerating Complex Queries on HDF\nDatasets Using Fast Bitmap Indices. In Proceedings of the\n18thInternational Conference on Scientific and Statistical Database\nManagement (SSDBM ’06). IEEE Com puter Society\, Washington\, DC\,\nUSA\, 149–158\n[14] Utkarsh Ayachit\, Andrew Bauer\, Berk Geveci\, Patrick O’Leary\,\nKenneth Moreland\, Nath an Fabian\,and Jeffrey Mauldin. 2015. ParaView\nCatalyst: Enabling In Situ Data Analysis and Visualization. In Proceedings\nof the First Workshop on In Situ Infrastructures for Enabling Extreme-Scale\nAnalysis and Visualiz ation (ISAV2015) . ACM\, New York\, NY\, USA\, 25–29\n\nhttps://events.c hpc.ac.za/event/47/contributions/1053/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1053/ END:VEVENT BEGIN:VEVENT SUMMARY:High Performance Data Access: the Hermes approach DTSTART;VALUE=DATE-TIME:20191202T090000Z DTEND;VALUE=DATE-TIME:20191202T093000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-956@events.chpc.ac.za DESCRIPTION:Speakers: Xian-He Sun ()\nHigh-performance computing (HPC) app lications generate massive amounts of data. However\, the performance impr ovement of disk-based storage systems has been much slower than that of me mory\, creating a significant I/O performance gap. To reduce the performan ce gap\, storage subsystems are under extensive changes\, adopting new tec hnologies and adding more layers into the memory/storage hierarchy. With a deeper memory hierarchy\, the data movement complexity of memory systems is increased significantly\, making it harder to utilize the potential of the deep memory-storage hierarchy (DMSH) architecture. In this talk\, we p resent the development of Hermes\, an intelligent\, multi-tiered\, dynamic \, and distributed I/O caching system that utilizes DMSH to significantly accelerate I/O performance. Hermes is a US NSF supported large software de velopment project. It extends HPC I/O stacks to integrated memory and para llel I/O systems\, extends the widely used Hierarchical Data Format (HDF) and HDF5 library to achieve application-aware optimization in a DMSH envir onment\, and enhances caching systems to support vertical and horizontal n on-inclusive caching in a distributed parallel I/O environment. We will in troduce the Hermes’ design and implementation\; discuss its uniqueness a nd challenges\; and present some initial implementation results.\n\nhttps: //events.chpc.ac.za/event/47/contributions/956/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/956/ END:VEVENT BEGIN:VEVENT SUMMARY:PETSc: High-Performance Software Library for Engineering and Scien ce Simulation DTSTART;VALUE=DATE-TIME:20191202T120000Z DTEND;VALUE=DATE-TIME:20191202T123000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-958@events.chpc.ac.za DESCRIPTION:Speakers: Hong Zhang (Mathematics and Computer Science Divisio n Argonne National Laboratory\, U.S.)\nPortable\, Extensible Toolkit for S cientific Computation (PETSc) is a suite of\ndata structures and routines for the scalable (parallel) solution of scientific applications. Due to it s solid mathematical grounding\, careful software design\, and most import antly\, evolution resulting from the usage of many users in various applic ation areas\, PETSc is enabling engineers and scientists to solve large sc ale problems\, with previously unreachable resolution\, in areas as divers e as groundwater contamination\, cardiology\, fusion\, nuclear energy\, as tro-physics\, and climate change.\n\nAs a PETSc developer\, I will give an overview of the PETSc\, and briefly introduce its basic use in algorithmi c research\, numerical production simulation and parallel performance eval uation. As an example\, I will present our recent simulation of the U.S. r iver systems on extreme-scale computers.\n\nhttps://events.chpc.ac.za/even t/47/contributions/958/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/958/ END:VEVENT BEGIN:VEVENT SUMMARY:Updating the Outdated Storage Paradigm to Handle Complex Computati onal Workloads DTSTART;VALUE=DATE-TIME:20191202T100000Z DTEND;VALUE=DATE-TIME:20191202T103000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-960@events.chpc.ac.za DESCRIPTION:Speakers: Craig Bungay (Spectra Logic)\nAccelerating discovery in computational science and high performance computing environments requ ires compute\, network and storage to keep pace with technological innovat ions. Within a single organization\, interdepartmental and multi-site shar ing of assets has become more and more crucial to success. Furthermore\, a s the growth of data is constantly expanding\, storage workflows are excee ding the capabilities of the traditional filesystem. For most organization s\, facing the challenge of managing terabytes\, petabytes and even exabyt es of archive data for the first time can force the redesign of their enti re storage strategy and infrastructure. Increasing scale\, level of collab oration and diversity of workflows are driving users toward a new model fo r data storage. \n\nIn the past\, data storage usage was defined by the te chnology leveraged to protect data using a pyramid structure\, with the to p of the pyramid designated for SSD to store ‘hot’ data\,’ SATA HDDs used to store ‘warm’ data and tape used for the bottom of the pyramid to archive ‘cold’ data. Today\, modern data centers have moved to a n ew two-tier storage architecture that replaces the aging pyramid model. Th e new two-tier paradigm focuses on the actual usage of data\, rather than the technology on which it resides. The new two-tier paradigm combines a p roject tier that is file-based and a second or perpetual tier which is obj ect based. The object based perpetual tier includes multiple storage media types\, multi-site replication (sharing)\, cloud\, and data management wo rkflows. Data moves seamlessly between the two tiers as data is manipulate d\, analyzed\, shared and protected – essentially creating yin and yang between the two storage tiers. Solutions designed to natively use the Perp etual Tier empower organizations to fully leverage their primary storage i nvestments by reducing the overall strain on the Primary Tier\, while at t he same time\, enabling data centers to realize numerous benefits of the P erpetual Tier that only increase as the amount of storage to manage increa ses.\n\nThe next logical question is how to manage data between the two ti ers while maintaining user access and lowering overall administration burd ens. Join us for a deeper look into the nuances of the two-tier system and data management between them. We will cover storage management software o ptions\; cloud vs. on-premise decisions\; and using object storage to expa nd data access and create a highly effective storage architecture to break through data lifecycle management barriers\n\nhttps://events.chpc.ac.za/e vent/47/contributions/960/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/960/ END:VEVENT BEGIN:VEVENT SUMMARY:Designing Reusable Composable Components for the (HPC) I/O Stack DTSTART;VALUE=DATE-TIME:20191202T113000Z DTEND;VALUE=DATE-TIME:20191202T120000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-957@events.chpc.ac.za DESCRIPTION:Speakers: Ali Butt (Virginia Tech)\nThe rise of AI/ML in HPC a pplications is also driving the need for suitable storage abstractions suc h as the key-value (KV) stores. These abstractions pose new challenges for the HPC I/O stack. Enterprise KV stores are not well suited for HPC appli cations\, and entail customization and cumbersome end-to-end KV design to extract the applications needs. To this end\, I will present BESPOKV\, an adaptive\, extensible\, and scale-out KV store framework. BESPOKV decouple s the KV store design into the control plane for distributed management an d the data plane for local data store. BESPOKV takes as input a single-ser ver KV store\, called a datalet\, and transparently enables a scalable and fault-tolerant distributed KV store service. The resulting distributed st ores are also adaptive to consistency or topology requirement changes and can be easily extended for new types of services. I’ll show that BESPOKV -enabled distributed KV stores scale horizontally to a large number of no des\, and performs comparably and sometimes better than the state-of-the-a rt systems.\n\nhttps://events.chpc.ac.za/event/47/contributions/957/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/957/ END:VEVENT BEGIN:VEVENT SUMMARY:Smoke and ventilation simulation using the CHPC DTSTART;VALUE=DATE-TIME:20191204T092000Z DTEND;VALUE=DATE-TIME:20191204T094000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1061@events.chpc.ac.za DESCRIPTION:Speakers: Kenneth Allen (Greenplan Consultants)\nKenneth Allen - Greenplan Consultants - kennethguyallen@greenplan.co.za \n\nGreenpl an Consultants undertook a combined smoke\, ventilation and wind study for a basement/underground parking area of approximately 27 000 m$^2$. This p resentation gives an overview of the project and the experience of running it at the CHPC. In order to reduce the size of the transient model\, the full domain (about 0.25 km$^3$) was modelled under steady-state condition s using OpenFoam. The wind flow patterns around the basement were then imp osed as boundaries on the transient model of the basement. Even with this approach\, the transient model was too large to simulate on a small office network\, so high performance computing (HPC) was essential. \n\nFire Dyn amics Simulator (FDS) 6.7.0. was used for the transient simulations. It is purpose-written for simulating fire and smoke\, and uses the computationa lly intensive Large Eddy Simulation (LES) method. The FDS model CHPC requi rements were as follows - RAM: ≈ 100 GB\; Nodes used: 10-15\; Cores used : 240-360\; Simulated time per case: ≈ 1-2 min\; Wall time per case: ≈ 20-60 hours\; Total cells: max. 70-80 million\; Cell size: 100 mm\; Data output per simulation: 200-250 GB. \n\nFDS can make use of OpenMP (Multi-P rocessing) and MPI (Message Passing Interface). Tests on the CHPC with Ope nMP enabled showed little-to-no improvement\, so OpenMP was set to 1 (disa bled). An experiment was made whereby all cores per node were booked on PB S but only half were used to run MPI process. This did not give better per formance – possibly because of ghost processes running on one or two of the cores. Subsequently\, all models were run with 1 MPI process per core\ , with 24 MPI processes per node. This seemed to be the best option for co st-effective performance. \n\nFDS makes use of manually-specified rectilin ear meshes. Unfortunately\, where parallel simulation is desired\, each MP I process requires at least one mesh\, which means the mesh domain must be manually split up into sub-meshes. Any mismatch in mesh size leads to une ven loading on the cores\, which means that the less heavily loaded cores have to wait. Due to the rectilinear grid\, there are also “wasted” ce lls in walls/floors/roofs which have no function but contribute to the com putational load. Thus\, although the initial CSIR scaling tests on the CHP C (6 million cells) were promising for cell counts as low as 30 000 cells per core\, scaling tended to be less efficient than expected. \n\nOn a num ber of occasions the simulations progressed at different speeds despite ha ving the same configuration\, flow speeds\, and boundary conditions. It is possible that this might have been caused by ghost processes on individua l cores and the CHPC architecture effect – in particular\, the blocking ratio between racks. As an experiment on our final model geometry\, we red uced the number of nodes in use from 15 (360 cores) to 10 (240 cores). The cell count per core was a factor of 2.5 higher\, the area modelled was la rger\, and there were more jet fans and extraction fans than in the previo us model. Despite this\, the CHPC wall time required per unit simulated ti me increased by a factor of only 2. Rigorous testing is necessary before a ny conclusions are drawn\, as a rough test like this does not provide suff icient data and there might be factors not taken into account. While the above meant that simulations did not run as fast as desired\, they ran far faster than they would have on a small office network (it would have take n years\, if they ran at all). Greenplan had a good experience with the CH PC and are keen to use it for future projects of this nature.\n\nhttps://e vents.chpc.ac.za/event/47/contributions/1061/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1061/ END:VEVENT BEGIN:VEVENT SUMMARY:High throughput in silico screening for tailored catalytic reactiv ity and selectivity DTSTART;VALUE=DATE-TIME:20191204T090000Z DTEND;VALUE=DATE-TIME:20191204T092000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1060@events.chpc.ac.za DESCRIPTION:Speakers: Caroline M. Krauter (Schrödinger GmbH)\nFirst-princ iples simulation has become a reliable tool for the prediction of structur es\, chemical mechanisms\, and reaction energetics for the fundamental ste ps in homogeneous and heterogeneous catalysis. Details of reaction coordin ates for competing pathways can be elucidated to provide the fundamental u nderstanding of observed catalytic activity\, selectivity\, and specificit y. Such predictive capability raises the possibility for computational dis covery and design of new catalysts with enhanced properties.\n\nIn the cas e of mesoporous materials like zeolites\, the well-defined pore structures and adjustable reactivity centers in the pore walls allow for efficient c ontrol of the catalytic properties. In addition to the reactivity at the c atalytic center\, the mobility of the reaction components throughout the n etwork structure is crucial to the design. In this contribution we will us e GPU-accelerated molecular dynamics simulations to study the diffusion of small molecules through zeolite structures.\n\nhttps://events.chpc.ac.za/ event/47/contributions/1060/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1060/ END:VEVENT BEGIN:VEVENT SUMMARY:NEXTGenIO: exploring the potential of non-volatile memory in HPC DTSTART;VALUE=DATE-TIME:20191202T123000Z DTEND;VALUE=DATE-TIME:20191202T130000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-961@events.chpc.ac.za DESCRIPTION:Speakers: David Homan (EPCC)\nMemory and storage read and writ e costs can lead to a significant loss of time and energy in current HPC s ystems. Byte-addressable non-volatile memory (NVM) could provide considera ble improvements in both time and energy requirements over conventional DR AM memory. Using Optane DCPMM\, Intel's new byte-addressable and persisten t memory\, the NEXTGenIO project investigated the performance of NVRAM by designing\, building and testing a bespoke prototype NVM system. The main goal of the project was to explore the potential of NVRAM in overcoming pe rformance bottlenecks in I/O and main memory\, which are considered signif icant barriers to Exascale computing.\n\nIn this talk we will give a brief overview of the NEXTGenIO system (192GB DRAM and 3TB of NVM per dual sock et node)\, and the various NVRAM usage modes. The results from a number of investigative test cases run on the NEXTGenIO prototype system will be pr esented. In particular we will discuss I/O performance\, run-time\, and en ergy consumption for applications with large I/O demands\, such as OpenFOA M and CASTEP. Comparison of the results from NVRAM and DRAM shows that NVR AM can indeed provide significant improvement in both performance and ener gy consumption.\n\nhttps://events.chpc.ac.za/event/47/contributions/961/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/961/ END:VEVENT BEGIN:VEVENT SUMMARY:Evaluating the Compressibility of Elevation Data using Space-Filli ng Curves DTSTART;VALUE=DATE-TIME:20191203T100000Z DTEND;VALUE=DATE-TIME:20191203T103000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1150@events.chpc.ac.za DESCRIPTION:Speakers: Conrad Haupt (University of the Witwatersrand)\nTwo of the typical points of interest with elevation data\, or Geographic Info rmation Systems\n(GIS) data in general\, are storage and query costs. The former is typically addressed by\nintegrating standard compression schemes into already existing storage mechanisms\, such\nas GZIP in HDF5. Space-F illing Curves (SFCs) have already been used to reduce access\ntime for spa tial operations on point and polygon data. In this research\, we evaluate the effect\nof using SFCs as a pre-processing step for standard compressio n schemes on elevation\ndata. We break up common compression tools into th eir base algorithms and identify\ncanonical SFCs from the literature (for example\, the Hilbert curve).\n\nWe use 1-arcsecond resolution elevation m aps from the Shuttle Radio Topographic Mission\n(SRTM) as the comparative data-set upon which we apply all combinations of SFCs and\ncompression sch emes. The SFCs\, in most cases\, neither significantly improve nor worsen\ ncompression ratios when compared to non-preprocessed results. However\, w e show that\ncertain pre-processing steps improve the compression performa nce of otherwise ineffective\ncompression techniques. This research shows the potential for future work on compression\nschemes which allow for in-p lace search and modifications without the loss of compression\nperformance . Another application is to apply these techniques to astronomical data fr om the\nSquare-Kilometre Array\, a major scientific and engineering projec t in South Africa\, for which\nsome preliminary results have been attained .\n\nhttps://events.chpc.ac.za/event/47/contributions/1150/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1150/ END:VEVENT BEGIN:VEVENT SUMMARY:KEYNOTE 1: Digital Transformation: Issues of the Public Service DTSTART;VALUE=DATE-TIME:20191202T074500Z DTEND;VALUE=DATE-TIME:20191202T083000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1169@events.chpc.ac.za DESCRIPTION:Speakers: Mandla Ngcobo ()\nGenerally\, a lot of awareness has been created around Digital Transformation/ the Fourth Industrial Revolut ion (4IR) with various stakeholders electing to focus on matters of intere st to them. However\, there are known challenges and requirements for the public service to be able to transform digitally. The presentation would t alk about challenges/ issues (people/ process/ technology issues) that req uire interventions/ solutions to achieve a Digitally Transformed public se rvice.\n\nhttps://events.chpc.ac.za/event/47/contributions/1169/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1169/ END:VEVENT BEGIN:VEVENT SUMMARY:Research highlights from the Molecular Bio-computations and Drug D esign Research Group at UKZN DTSTART;VALUE=DATE-TIME:20191202T113000Z DTEND;VALUE=DATE-TIME:20191202T115000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1165@events.chpc.ac.za DESCRIPTION:Speakers: Mahmoud soliman (Molecular Bio-computation and Dru g Design Laboratory\, School of Health Sciences\, University of KwaZulu-N atal\, Westville Campus\, Durban 4001\, South Africa)\nProf Mahmoud Solima n (http://soliman.ukzn.ac.za/) will provide a presentation that highlights the various research scopes and outcomes of research that is being conduc ted in his research group at UKZN over the last 8 years with more emphasis on the applications of computational simulations and the contribution of CHPC in drug design and discovery as well as capacity development. A few s elected research topics will be presented such as: the irony of chirality\ ; covalent drug inhibition\; does drug size matter\; and other topics.\n\n https://events.chpc.ac.za/event/47/contributions/1165/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1165/ END:VEVENT BEGIN:VEVENT SUMMARY:More than meets the eye: Towards an Artificial Intelligence Observ atory DTSTART;VALUE=DATE-TIME:20191204T090000Z DTEND;VALUE=DATE-TIME:20191204T092000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1162@events.chpc.ac.za DESCRIPTION:Speakers: Matias Carrasco Kind (NCSA)\nFile attached\n\nhttps: //events.chpc.ac.za/event/47/contributions/1162/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1162/ END:VEVENT BEGIN:VEVENT SUMMARY:Cloud Computing a Solution to Internet Services Downtime at Namibi an Institutions of Higher Learning DTSTART;VALUE=DATE-TIME:20191203T090000Z DTEND;VALUE=DATE-TIME:20191203T092500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1148@events.chpc.ac.za DESCRIPTION:Speakers: Shadreck Chitauro (Namibia University of Science and Technology)\nE-learning has been adopted by many institutions as a means by which to foster an effective teaching and learning environment. Namibia n institutions of higher learning have not been left behind and are also u sing e-learning systems. Use of e-learning brings about many benefits like that of students being able to access learning material anywhere and anyt ime but implementation of e-learning has its own drawbacks of which cloud computing could be the answer. Furthermore\, higher learning institutions run their own IT systems and buy their own IT infrastructure which has imp lications on the overall institutions’ budgets. Data on issues with over all IT operations not just for e-learning systems was obtained from a case site at one of the Namibian institutions of higher learning. The study sh owed that the institutions’ IT systems have problems that cloud computin g could be able to solve. \n\nKeywords: Cloud computing\, e-learning\, edu cation\, downtime\n\nhttps://events.chpc.ac.za/event/47/contributions/1148 / LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1148/ END:VEVENT BEGIN:VEVENT SUMMARY:KEYNOTE 3 (SA NREN): eInfrastructures as enablers for global scien ce and education DTSTART;VALUE=DATE-TIME:20191203T070000Z DTEND;VALUE=DATE-TIME:20191203T074500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1139@events.chpc.ac.za DESCRIPTION:Speakers: Josva Kleist (NORDUnet A/S)\nScience and education a re increasingly becoming global endeavors. Scientists collaborate internat ionally and questionnaires indicate that scientists today are more likely to collaborate with partners outside their home institutions than with loc al research fellows. Students no longer need to physically be on campus to take classes\, as they follow courses remotely and can even take a degree at foreign university without leaving home. Furthermore\, there is a grow ing number of international research infrastructures\, some centralized an d some highly distributed\, in all areas of science. The construction of t hose infrastructures is partly driven by financial necessity\, but also fr om a need to bring together the necessary competences. Furthermore\, the p aradigms of Open Science and Citizen Science foster foster an environment of sharing and inclusion. Collaboration is no longer just a matter of scie ntist visiting for weeks and months.\n \nFor these reasons\, access to fac ilities\, research infrastructures\, and data must be global\, universal\, fine-grained\, and instant. As an eInfrastructure provider\, R&E Network organizations are one of the fundamental building blocks supporting today ’s global science\, and just as scientific activities are becoming more and more global\, we need to think global. With the network\, we already s ee this happening\, where leading NRENs\, such as the South African NREN\, participate in the Global Network Advancement Group\, that works on the i ntercontinental aspects of the (GREN) Global R&E Network. Moreover\, netwo rk services such as eduroam\, eduGAIN\, and eduVPN take a truly global app roach. The infrastructure needed to support Open Science and Citizen Scie nce drives us to break down silos between storage\, compute\, and network infrastructures. We need to think about it in its entirety. The same goes for some of the large international research infrastructures\, where stora ge\, compute\, and the network become an integral part of the science inst rument. In this talk\, I will present my views on the role of R&E Network s as enablers for science and education\, with an outset on what I see hap pening in the European Nordics\, at a European level\, and globally.\n\nht tps://events.chpc.ac.za/event/47/contributions/1139/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1139/ END:VEVENT BEGIN:VEVENT SUMMARY:Cybers Security Threats – Is Our Research Data Safe? DTSTART;VALUE=DATE-TIME:20191204T123000Z DTEND;VALUE=DATE-TIME:20191204T125000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1138@events.chpc.ac.za DESCRIPTION:Speakers: Isak van der Walt (University of Pretoria)\nIn a wor ld where more and more people are connected every day\, new systems pop up and\naccess can be gained from any device anywhere\, we need to ask if th e data that we are working\non (viewing\, editing and manipulating) is ori ginal and safe? The value of data as an asset is\nincreasing as well as th e potential threat in its transformative and informative power. The concep t\nand roll out of Institutional Repositories (IR) and Institutional Data Repositories (IDR) is fairly\nmainstream in South Africa\, hosting large v aluable and often invaluable sets of data. In a\nconnected world\, cyber t hreats are real and have to date caused harm to many organisations. Is\nou r data really safe and trustworthy in South Africa? South Africa has seen a dramatic increase\nin the amount of cyber-attacks in 2019\, with as much as a 22% increase in malware attacks\ncompared to 2018 alone. Malware att acks only form part of an array of cyber threats that are\ncarried out on a daily basis. In short\, the answer is that no system is 100% safe agains t cyber\nattacks\, there are however systems and processes in place to inc rease and safeguard our data.\n\nToday’s cybercriminal strategies target every link in the attack chain to gain access to resources\nand data\, ex ploiting them relentlessly. Holding data for ransom\, modifying data\, def acing of data\nand selling of data are some of the perils that more and mo re organisations are facing when\nbreached. The key to safeguarding data i s to ensure that policies\, systems and procedures are\nput in place to de al with the various links in the attack chain. This presentation will focu s on what\nis happening on an international and national level with regard s to cyber security threats and how\nit affects our research data. The pre sentation will also highlight the attack chain and what can be\ndone at th e various linkages to make systems and organizations more secure.\n\nhttps ://events.chpc.ac.za/event/47/contributions/1138/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1138/ END:VEVENT BEGIN:VEVENT SUMMARY:Evolution of a South African eResearch support service DTSTART;VALUE=DATE-TIME:20191204T121000Z DTEND;VALUE=DATE-TIME:20191204T123000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1137@events.chpc.ac.za DESCRIPTION:Speakers: Dale Peters (UCT)\nINTRODUCTION\n\nGiven the complex relationship between eResearch stakeholders within universities and resea rch organisations\, a “one size fits all” solution to the development of a national eResearch support service is not practical. Services contri buted by Libraries\, IT and research administrations are at different stag es of maturity\, and the South African landscape is characterised by scarc e skills in areas of software and systems support. To advance national eRe search capability\, a considered approach requires both costly infrastruct ure investment and collaborative support services. \n\nBACKGROUND\n\nThe e mergence of a new paradigm\, “sometimes called eResearch”\, gave rise to the examination of a national information service framework in 2005.( ) The need for joint action was identified to meet the challenges of eResea rch cost-effectively in South Africa. A specialized agency was proposed to provide support services\, with a governance model that should work well for all participants. Two reports commissioned by the Department of Scien ce and Technology assisted in conceptualising strategic plans for the furt her development of South Africa’s research infrastructure\, including th e cyberinfrastructure component. The recommendation to establish of a Nat ional Integrated Cyberinfastructure System (NICIS) was accepted and plans for follow-up activities approved in 2013. NICIS comprises several core components of the Tier 1 infrastructure: a national Center for High Perfor mance Computing (CHPC)\, the South African Research Network (SANReN)\, and the more recently established Data Intensive Research Initiative for Sout h Africa (DIRISA). Experience of the Research Data Management project com ponent of the DIRISA Tier 2 ilifu infrastructure has provided valuable les sons the collaborative development of shared services that can now be evol ved to the wider community. \n\nFROM INFRASTRUCTURE DEVELOPMENT TO SERVIC E ORIENTATION\n\nAs research becomes more multidisciplinary\, more collabo rative and more global\, researchers seek to leverage the South African in vestment in specialist scientific equipment and domain-specific infrastruc tures\, often generating massive data outputs for analysis in internationa l collaboration. As the national research infrastructure moves from an ex perimental testbed to a user-oriented environment\, a challenge faced by m ost eResearch infrastructures is the provisioning of sustainable services\ , and the monitoring of user experience (UX)\, to improve the interaction of researchers with the infrastructure. This critical component is seldom defined explicitly in the infrastructure development\, and the research c ommunity have little interest in the expansion of cost-effective services beyond their own needs\, and especially beyond the duration of their funde d project. Responsibility at present\, falls to the host entity to realise the full potential of the national cyberinfrastructure\, and the collabor ation enabled with global infrastructures. A limited science system sugge sts a federation of distributed support services\, including multiple univ ersities and institutional partners to meet the ever-increasing need to me et both current user support and ongoing data access.\n\nA pilot project t o support a South African eResearch support service will build on the eRes earch Africa conference hosted bi-annually at the University of Cape Town since the initial event in 2013. An annual training workshop aimed at pro fessional development and career enhancement opportunities recognizes the varied job roles associated with eResearch. Institutional eResearch capaci ty building will focus on selected teams of information professionals thro ugh sponsored participation in designated training programmes and national events. \n\nCONCLUSIONS\n\nThe development of a national support service model is intended to improve distributed efficiency\, rather than to centr ally consolidate a limited pool of existing human resources. The effect o f overextending the existing capacity poses serious threat to the realisat ion of the national cyberinfrastructure\, with discussion of actual use ca ses in this presentation. \n\nDue to the complex relationship between eRes earch stakeholders within institutions\, a “one size fits all” solutio n is impractical\, and a phased approach is recommended\, leveraging a bro kerage model to access third party services and avoid scenarios where serv ices are developed and implemented and then subsequently “orphaned” by lack of support and changing financial priorities. \n\nThe potential admi nistrative overhead of service development projects\, established by indiv idual service level agreements with multiple institutions\, warrants furth er consultation on the project governance with university executives\, sen ior researchers and infrastructure managers. \n\nCapability approaches to advanced computing technologies must address more than the big shiny stuf f. A considered approach requires both costly infrastructure investment an d collaborative support services. The user experience of researchers\, and their improved interaction with the national cyberinfrastructure should u ltimately direct the project and its evaluation.\n\nhttps://events.chpc.ac .za/event/47/contributions/1137/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1137/ END:VEVENT BEGIN:VEVENT SUMMARY:Developments Around Data in Botswana to support 4iR preparedness\, Research\, Innovation in a Developing Country DTSTART;VALUE=DATE-TIME:20191204T115000Z DTEND;VALUE=DATE-TIME:20191204T121000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1136@events.chpc.ac.za DESCRIPTION:Speakers: Tshiamo Motshegwa (University Of Botswana)\nThe 3rd Industrial revolution of the 20th Century ushered in the 1st Information R evolution that brought the internet\, digitisation\, digitilisation and di gital transformation and created a basis for knowledge-based economies. It is now widely accepted that the world is experiencing the advent of the 2 nd Information revolution that is ushering in a 4th Industrial Revolution - a revolution that is characterised by a fusion of technologies to addre ss current and future human needs. The 4th Industrial revolution is also c haracterised by large amounts and variety of data – coming from various sources at high frequency – and our ability to analyse them\, in real-t ime and derive information and knowledge for timely decision making. It is anticipated that the 4th Industrial revolution will revolutionise Industr y production processes through advanced automation (often referred to Ind ustry 4.0)\, it also further anticipated that it will revolutionise pract ice and effectiveness in a variety of areas - health care provision (e.g. personalised medicine)\, development of Smart Cities \, precision agricul ture and help address weather and climate change. \n\nThere is therefore n eed for African countries and developing countries to respond to the onset on the 4th Industrial revolution (amidst arguably addressing challenges f rom previous industrial revolutions still unravelling in the continent) . This will help to bridge the digital divide and help not leave anyone behi nd – and achieve Africa’s vision 2063 – the Africa We Want. It will also help accelerate attainment of Sustainable Development goals through r iding on technology advances\, efficiency and transparency. There is need for Africa to transform its infrastructure\, research & innovation ecosys tems\, skills and education systems etc – this for 4th Industrial revolu tion readiness and competitiveness of African economies and Africans in th is new dispensation. Africa and African countries need to have tailor made responses to the 4th Industrial revolution and its implications to the Af rican context. This can be done by concretising National\, Regional and Co ntinental Policy Frameworks\, structures\, resourced roadmaps and increase d expenditure in Research\, Science\, Technology and Innovation and develo ping partnerships . The universities and other centers of knowledge creati on and skills development must play a critical role. The Universities must be alive to this responsibility and aim to transform to be a research-int ensive institutions – and is enhance their internal university innovati on ecosystems including around data exploitation through innovations.\n\nT his talk will provide an update on the developments of around data in Bots wana\, this from the prism of policy and strategy development\, research a nd Innovation\, skills development and science communication for public an d policy engagement – this to help address Botswana’s socio-economic c hallenges and attainment of Vision 2036 – Prosperity for all and address 4IR preparedness. The talk will discuss developments around Botswana Open Data Open Science\, Botswana Space Science and Technology Strategy Develo pment\, highlight example National Open Data projects and discuss the Univ ersity Industry Government Co-creation Initiative that aims to foster inno vation- including around exploitation of open data.\n\nhttps://events.chpc .ac.za/event/47/contributions/1136/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1136/ END:VEVENT BEGIN:VEVENT SUMMARY:Developing infrastructure for federated data analysis for protecte d human data DTSTART;VALUE=DATE-TIME:20191204T113000Z DTEND;VALUE=DATE-TIME:20191204T115000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1135@events.chpc.ac.za DESCRIPTION:Speakers: Nicola Mulder (University of Cape Town)\nDatasets of ever increasing size and complexity are being generated in the biomedical field to answer questions about human and animal health. Data on human he alth have to be managed responsibly to ensure protection of participants i n health studies. Additionally\, many governments are clamping down on the transfer of datasets out of country borders. In order to respect these co ncerns while still facilitating ethical and responsible data sharing for a nalysis\, new policies and infrastructure need to be developed. There are several initiatives working in this space\, including the Global Alliance for Genomics and Health (GA4GH)\, which is building standards and tools fo r sharing of genomic data globally. A new EU funded research project\, CIN ECA (Common Infrastructure for National Cohorts in Europe\, Canada\, and A frica)\, is developing infrastructure to implement GA4GH standards to enab le the analysis of data across cohorts without the requirement for the tra nsfer of large datasets to third parties. This includes development of sec urity systems for authentication and authorization of researchers\, harmon ization of data across heterogenous studies\, and development of cloud-bas ed tools for federated data analysis within the confines of participant co nsent. This presentation will describe some of the standards and tools bei ng developed and implemented in the CINECA project.\n\nhttps://events.chpc .ac.za/event/47/contributions/1135/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1135/ END:VEVENT BEGIN:VEVENT SUMMARY:Federating the Ilifu Cloud DTSTART;VALUE=DATE-TIME:20191204T100000Z DTEND;VALUE=DATE-TIME:20191204T102000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1134@events.chpc.ac.za DESCRIPTION:Speakers: Rob Simmonds (University of Cape Town)\nIlifu is an Infrastructure as a Service cloud that utilises OpenStack to provide its c ore services. It is run by a consortium of South African universities and provides data intensive computing resources to Astronomy and Bioinformatic s users. This talk describes how we are utilising federated identity servi ces enable the use of ilifu by users in a way that can be managed by indiv idual project groups without needing to contact ilifu support to have them create or remove accounts. It will explain how this has been done using t ools run by EGI\, and using tools integrated locally.\n\nhttps://events.ch pc.ac.za/event/47/contributions/1134/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1134/ END:VEVENT BEGIN:VEVENT SUMMARY:The Capture\, Storage and Consumption of the MeerKAT Radio Telesco pe’s Sensor Data DTSTART;VALUE=DATE-TIME:20191204T094000Z DTEND;VALUE=DATE-TIME:20191204T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1133@events.chpc.ac.za DESCRIPTION:Speakers: Suleiman Hoosen (South African Radio Astronomy Obser vatory)\nThe Control and Monitoring (CAM) Team at the South African Radio Astronomy Observatory (SARAO) is responsible for implementing software sol utions for the collection of sensor data from all of the components\, user -supplied equipment and ancillary devices that make up the 64-dish MeerKAT Radio Telescope. Recently\, the CAM Team developed and deployed a new sol ution called KatStore64 which provides services to the MeerKAT Telescope O perators\, Astronomers and academia to access the telescope’s sensor dat a. \nIn order to capture\, store and consume all of this data\, the CAM Te am makes use of the services offered by the Centre for High Performance Co mputing (CHPC) for long term storage of the data and employs APIs to extra ct and present the stored data to users. \n\nMy poster/talk will illustrat e how the CAM Team has built and implemented the KatStore64 service for th e storage and retrieval of sensor data for the MeerKAT Radio Telescope.\n\ nhttps://events.chpc.ac.za/event/47/contributions/1133/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1133/ END:VEVENT BEGIN:VEVENT SUMMARY:EOS storage system DTSTART;VALUE=DATE-TIME:20191204T092000Z DTEND;VALUE=DATE-TIME:20191204T094000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1132@events.chpc.ac.za DESCRIPTION:Speakers: Sean Murray (CHPC\, CSIR)\nEOS is the storage system of choice for CERN's data storage and\nis used by various WLCG Tier1 and Tier2 facilities\, including at the CHPC.\nEOS is an elastic\, adaptable\, and scalable software based solution for\ncentral data recording\, user a nalysis and data processing.\n\nIt has a multitude of supported protocols and authentication methods.\nWe will present what EOS is\, what is does\, and how we use EOS in \nconjunction with our Tier 2 facility\, and how EOS is used in a couple of \nother examples.\n\nhttps://events.chpc.ac.za/eve nt/47/contributions/1132/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1132/ END:VEVENT BEGIN:VEVENT SUMMARY:Next Generation Sequencing: where big data and high-performance co mputing meet DTSTART;VALUE=DATE-TIME:20191202T092000Z DTEND;VALUE=DATE-TIME:20191202T094000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1045@events.chpc.ac.za DESCRIPTION:Speakers: Brigitte Glanzmann (DST-NRF Centre of Excellence for Biomedical Tuberculosis Research\, South African Medical Research Council Centre for Tuberculosis Research\, Division of Molecular Biology and Huma n Genetics\, Faculty of Medicine and Health Sciences\, Stellenbosch Univer sity\, Cape Town\, South Africa. )\nThe advent and evolution of next gener ation sequencing (NGS) has considerably impacted genomic research\, includ ing precision medicine. High-throughput technology currently allows for th e generation of billions of short DNA or RNA sequence reads within a matte r of hours. This becomes extremely important in the case of genetic disord ers where rapid and inexpensive access to a patient’s individual genomic sequence is imperative and enables target variant identification. NGS tec hnologies results in the generation of large data sets which require exten sive bioinformatic and computational resources. Computational life science s therefore relies on the implementation of well-structured data analysis pipelines as well as high-performance computing (HPC) for large-scale appl ications. Here\, we report the sequencing of the first six whole human gen omes in South Africa and the processing of the data in collaboration with the Centre for High Performance Computing (CHPC). Efficient parallel and d istributed implementations of common time-consuming NGS algorithms on mode rn computational infrastructures are imperative. The latter becomes pivota l as NGS will continue to transcend from research labs to clinical applica tions in the near future.\n\nhttps://events.chpc.ac.za/event/47/contributi ons/1045/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1045/ END:VEVENT BEGIN:VEVENT SUMMARY:Parallel and Distributed Search Algorithms DTSTART;VALUE=DATE-TIME:20191203T090000Z DTEND;VALUE=DATE-TIME:20191203T093000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1041@events.chpc.ac.za DESCRIPTION:Speakers: Cornelia Inggs (Stellenbosch University)\nWe investi gate the parallelisation and performance analyses of search and planning a lgorithms for artificial intelligence\, machine learning\, and software ve rification. These applications involve the exploration of large state spac es\, which requires at its core a combinatorial search algorithm. Much of our work\, therefore\, focuses on evaluating and improving the scalability of algorithms used in all these tasks.\n\nIn recent work we have implemen ted various parallel and distributed MCTS algorithms with different enhanc ement strategies for artificial intelligence\, tested them for scalability \, and compared the performance of these approaches on the same domain and the same hardware infrastructure. We make use of the CHPC's large queue to determine scalability up to 128 12-core compute nodes with 32GB RAM eac h---values that are in line with previous publications and distributed sea rch implementations. We wrote our application code in Java\, using an acto r model framework (Akka) to simplify concurrency and distributed computing . We make limited use of MPI---more specifically\, just mpirun---in order to easily launch our application on the available nodes using the PBS node file.\n\nThis talk will provide an overview of our research and the proble ms we investigate\, as well as a discussion of recent results.\n\nhttps:// events.chpc.ac.za/event/47/contributions/1041/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1041/ END:VEVENT BEGIN:VEVENT SUMMARY:Advancing the understanding of interactions between plasma arcs an d power electronics with high performance computing DTSTART;VALUE=DATE-TIME:20191203T094000Z DTEND;VALUE=DATE-TIME:20191203T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1054@events.chpc.ac.za DESCRIPTION:Speakers: Quinn Reynolds (Mintek)\nDirect-current (DC) arc fur naces account for a significant proportion of installed pyrometallurgical capacity worldwide. Their applications include steel recycling as well as smelting of various materials such as ferrochromium\, ferronickel\, ilmeni te\, and others. In order to provide power to such furnaces\, alternating current from the grid or other generation sources must be converted into D C by rectification. At industrial scales the rectifier unit is often the s ingle largest capital cost item\, and any errors in its specification can result in the entire plant operating inefficiently (or not at all).\n\nIn this presentation\, computational plasma arc models developed in OpenFOAM ® are coupled with circuit simulations of solid-state furnace rectifiers in order to gain insight into the complex interactions between the rectifi er’s design parameters and the behaviour of the arc. Such approaches pro vide a first step toward true virtual prototyping and digital twin modelli ng for the electrical design and optimisation of DC arc furnaces. \n\nHigh performance computing is a critical enabling tool in such studies\, and v arious aspects of this – including solver performance scaling analysis\, software automation\, and use of methodologies from other HPC fields – will be touched on during the presentation.\n\nhttps://events.chpc.ac.za/e vent/47/contributions/1054/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1054/ END:VEVENT BEGIN:VEVENT SUMMARY:Computational Modelling Study on Stability of Li-S/Se System DTSTART;VALUE=DATE-TIME:20191204T121000Z DTEND;VALUE=DATE-TIME:20191204T123000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1067@events.chpc.ac.za DESCRIPTION:Speakers: Masedi Cliffton (UL)\nLithium Sulphur batteries suff ers from the low conductivity of S and the solubility of intermediary poly sulfide species during cycling. It has been reported that Se and mixed Sex Sy represent an attractive new class of cathode materials with promising e lectrochemical performance in reactions with both Li ions. Notably\, unlik e existing Li/S batteries that only operate at high temperature\, these ne w Se and Li/SexSy electrodes are capable of room temperature cycling. To s tudy large systems and impact of temperature effectively\, empirical inter atomic potentials of Li2S were derived and validated against available exp erimental structure and elastic properties. Complex high temperature trans formations and melting of Li2S was reproduced\, as deuced from molecular d ynamics simulations. Li2S was found to withstand high temperatures\, up to 1250K each which is a desirable in future advanced battery technologies. Cluster expansion and Monte-Carlo simulations were employed to determine p hase changes and high temperature properties of mixed Li2S-Se. The former generated 42 new stable multi-component Li2S-Se structures. Monte Carlo si mulations produced thermodynamic properties of Li2S-Se system for the enti re range of Se concentrations obtained from cluster expansion and it demon strated that Li2S-Se is a phase separating system at 0K but changes to mix ed system at approximately 350K.\n\nhttps://events.chpc.ac.za/event/47/con tributions/1067/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1067/ END:VEVENT BEGIN:VEVENT SUMMARY:Computational Catalysis @ NWU DTSTART;VALUE=DATE-TIME:20191204T113000Z DTEND;VALUE=DATE-TIME:20191204T115000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1064@events.chpc.ac.za DESCRIPTION:Speakers: Cornie van Sittert (North-West University)\nAt the N orth-West University (NWU) interest in incorporating computational chemist ry in training and research started in the late 1990. Although there was not much support for this interest in computational chemistry\, the need f or understanding the chemistry in especially catalysis was identified.\n\n Starting with old discarded computers and the cheapest possible software\, the first attempts were made in calculating structures of transition meta l complexes used in catalysis research at NWU. Due to the limitation of t he resources\, only gas phase reactions in homogeneous catalysis could be investigated. As the value of the computational investigations became evi dent\, more resources were acquired.\n\nIt was however only in 2002 that s upport from the Research Focus Area (previously called Separation Technolo gy) at NWU was obtained. Formal training of one staff member and the esta blishment of a dedicated Laboratory for Applied Molecular Modelling was fu nded. After careful evaluation of the needs in the research and the abili ties of the researchers\, it was decided to invest in Accelrys Materials S tudio (for research) and Spartan (for training) software. At the same tim e 10 workstations and a 12 CPU cluster were acquired.\n\nAlthough this was a major step forward\, catalysis research was still limited to gas phase reactions in homogeneous catalysis investigations\, with transition state calculations being a challenge. At this stage the CHPC was established. After a short phase of development and streamlining operations and softwar e at CHPC\, access to these resources were obtained by NWU researchers.\n\ nWith the access to CHPC resources\, limitations to the type of investigat ions gradually disappeared. The homogeneous catalysis investigations coul d be expanded to real system investigation\, including solvents. Models c ould be expanded from explanations of observations to prediction for activ ity. Heterogeneous catalysis could also be included in research.\n\nNow\, computational catalysis research at NWU was ready to investigate real pro blems and try to find solutions. One such real problem being investigated at NWU is the development of new/alternative catalysts to apply in the ge neration of alternative and renewable energy.\n\nhttps://events.chpc.ac.za /event/47/contributions/1064/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1064/ END:VEVENT BEGIN:VEVENT SUMMARY:Field-induced exotic electronic phases in spin-filter tunnel junct ions DTSTART;VALUE=DATE-TIME:20191202T094000Z DTEND;VALUE=DATE-TIME:20191202T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1046@events.chpc.ac.za DESCRIPTION:Speakers: Aniekan Ukpong (UKZN)\nABSTRACT:\nThe future of spin tronics based on 2D-materials is dependent on the effectiveness of the inj ection of pure spin current into a tunnel barrier region. Here\, first pri nciples calculations are used to show that the efficiency of the spin-filt ering across the semiconducting barriers of monolayer hBN is mainly limite d by the dynamical response of tunneling electrons to the applied axial fi eld. By projecting the effective electric field gradient densities and mag netic shielding constants across constitutive atomic layers in the scatter region of spin-filter tunnel junctions\, an unusual site-dependent spin r esponse is unraveled at the Fe/hBN and hBN/metal heterobilayer interfaces. Since the ground-state energy has no lower bound in extended electric fie lds\, our analyses of the dependence of the Fermi surface topology on appl ied electric fields show the emergence of a frustrated electronic order. T his exotic electronic phase is characterized by electric-field induced spi n-flip relative to the ferromagnetic ground state\, and observable as fiel d-tunable perpendicular magnetic anisotropy.\n\nHPC Content: \nAll the cal culations were performed in parallel using version 6.4.1` of the Quantum E SPRESSO suite. Due to poor code scalability\, all the computations were ca rried out on the 'SMP que' using 1 Node of 24 CPU cores. No net gain in co mputing speed was observed when more nodes were used on larger system size s. In fact\, the speed of the computations significantly reduced to > 6 CP U hours per scf-cycle when the same jobs were running on the 'Normal que' at 10 Nodes at double the system-size. The main computational challenge li es in solving the associated Poisson’s equation for atoms in the presenc e of the compensating potential due to externally-applied fields under gau ge-corrections\, in a more efficient manner. For fully-converged field-dep endent computations\, an average duration per task was timed at 2d 0h19m ( CPU time) and 2d 0h40m (WALL time). This is still too 'slow' for scientifi c computing jobs executed on a supercomputer.\n\nhttps://events.chpc.ac.za /event/47/contributions/1046/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1046/ END:VEVENT BEGIN:VEVENT SUMMARY:Convective scale modelling on the CHPC: ICON vs COSMO models DTSTART;VALUE=DATE-TIME:20191203T100000Z DTEND;VALUE=DATE-TIME:20191203T102000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1055@events.chpc.ac.za DESCRIPTION:Speakers: Patience Mulovhedzi (South African Weather Service)\ nThe South African Weather Service has recently embarked on a business cas e to address its computational needs. Part of this was to identify the mos t suitable convective scale Numerical Weather Prediction (NWP) model for t he Southern African region. The Unified model (UM)\, the main model run by SAWS for operational purposes\, the Weather Research and Forecasting (WRF ) Model and the Consortium for Small-scale Modeling (COSMO) model were use d for the study. A number of weather parameters were selected for the stud y\, and results generally showed that the three models are comparable. How ever\, with much model development taking place around the world\, the COS MO will soon be replaced by the Icosahedral Non-hydrostatic (ICON) model. It\, therefore\, makes sense to conduct the same study for the ICON as for the COSMO in order to investigate whether the new model is an improvement of the former one. Simulations for both the COSMO and ICON are run on the CHPC.\n\nhttps://events.chpc.ac.za/event/47/contributions/1055/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1055/ END:VEVENT BEGIN:VEVENT SUMMARY:In-silico investigations of metal coordinating enzymes: From Biofu el production to Antimicrobial drug resistance DTSTART;VALUE=DATE-TIME:20191204T094000Z DTEND;VALUE=DATE-TIME:20191204T100000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1062@events.chpc.ac.za DESCRIPTION:Speakers: Vuyani Moses (Vuyani)\nEnzymes which are directly bo und to metal cofactors are referred to as Metalloenzymes. These enzymes pl ay various biologically important roles from catalyzing electron transfer reactions to being important structural components of protein structures. Due to the abundance of metal containing enzymes and the role they play in important biological processes\, it is important to study these enzymes. In-sillico approaches are readily used to study protein structure and Mole cular Mechanics (MM) is an essential tool used for understanding protein d ynamics. MM is used to describe protein behavior by applying a MM force fi eld to describe bonded and nonbonded terms of a protein structure. The acc uracy of the force field in describing a particular protein structure is h ighly dependent on the force field parameters. Unfortunately\, for metallo enzymes there are no currently available force fields which can accurately describe the coordination environment of metals in metaloenzymes. As a re sult\, performing an accurate Molecular Dynamics (MD) for metalloenzymes i s extremely challenging using available force fields. To overcome this lim itation Quantum Mechanics (QM) may be applied to elucidate the parameters required for accurate description of metalloenzymes during MD simulations. This approach involves the use of potential energy surface (PES) scans to evaluate the angles\, bonds and dihedral parameters that are important to describe the metal binding site. Experimentally derived energy profiles g enerated from PES scans are then fitted using least squares fitting to a t heoretical force field to generate the force field parameters. This approa ch three cases of metal coordinating enzymes. The first are the Auxilliar y Activity family 9 (AA9) enzymes which are Cu(II) containing enzymes that have been shown to increase the rate of cellulose degradation. Secondly\, new parameters were also used in the identification of novel inhibitory c ompounds against the Mn(II) coordinating HIV-1 reverse transcriptase enzym e. Finally\, this approach was applied to the Zn(II) Bi metallic active si te center of Beta lactamase enzymes which are contributors to the developm ent of bacterial antibiotic resistance. For all three cases force field pa rameters were successfully generated and validated using MD simulations\n\ nhttps://events.chpc.ac.za/event/47/contributions/1062/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1062/ END:VEVENT BEGIN:VEVENT SUMMARY:Use of high-throughput sequencing to uncover resistance mechanisms in sugarcane DTSTART;VALUE=DATE-TIME:20191204T100000Z DTEND;VALUE=DATE-TIME:20191204T102000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1063@events.chpc.ac.za DESCRIPTION:Speakers: Robyn Jacob (South African Sugarcane Research Instit ute)\nThe sugarcane industry is an important agricultural activity in Sout h Africa generating an annual estimated average direct income of R14 billi on. Economic loss due to Eldana saccharina (eldana)\, a lepidopteran stem- borer\, is estimated to be R1 billion per annum. Commercial sugarcane cult ivars (Saccharum spp. hybrids)\, have different susceptibility ratings to eldana\, varying from low to high risk of sustaining economically damaging infestations. The South African Sugarcane Research Institute has utilised the resources of the Centre for High Performance Computing in an approach involving high-throughput RNA sequencing (RNA-seq) to identify early and late response genes that are differentially expressed in two sugarcane cul tivars possessing contrasting resistance phenotypes when challenged with e ldana herbivory. The results will be used to identify molecular mechanisms involved in the successful defence response and identify candidate genes which are most likely to be useful in breeding for resistance to eldana. \ n\nHPC content: \nAnnotation and assembly is a computationally intensive p rocess that requires considerable CPU time and effort. Various bioinformat ic tools were used for the de novo transcriptome assembly and the differen tial expression analyses required in this project.\n\nhttps://events.chpc. ac.za/event/47/contributions/1063/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1063/ END:VEVENT BEGIN:VEVENT SUMMARY:KEYNOTE 5 (DIRISA): DALI: A Data life cycle instrument for manage ment and sharing of data: Towards the reproducibility and data reuse of sc ientific research DTSTART;VALUE=DATE-TIME:20191204T070000Z DTEND;VALUE=DATE-TIME:20191204T074500Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1068@events.chpc.ac.za DESCRIPTION:Speakers: Hakizumwami Birali Runesha (University of Chicago)\n Researchers today are generating volume of data from simulations\, instrum ents and observations at accelerating rates\, resulting in extreme challen ges in data management and computation. In addition to publications\, scie ntists now produce a vast array of research products such as data\, code\, algorithms and a diversity of software tools. However\, scholarly publica tions today are still mostly disconnected from the underlying data and cod e used to produce the published results and findings\, which need to be sh ared. This presentation will discuss a funded project to acquire and opera te an extensible Data Lifecycle instrument (DaLI) for management and shari ng of data from instruments and observations that will enable researchers to (i) acquire\, transfer\, process\, and store data from experiments and observations in a unified workflow\, (ii) manage data collections over the ir entire life cycle\, and (iii) share and publish data. This presentation will also discuss our approach in generating and sharing data and artifac ts associated with research publications\, and therefore\, providing acces s to a platform that makes data findable\, accessible\, interoperable\, re usable and reproducible.\n\nhttps://events.chpc.ac.za/event/47/contributio ns/1068/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1068/ END:VEVENT BEGIN:VEVENT SUMMARY:Speech data harvesting and factorized deep neural network developm ent for the low-resource indigenous languages of South Africa DTSTART;VALUE=DATE-TIME:20191204T115000Z DTEND;VALUE=DATE-TIME:20191204T121000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1065@events.chpc.ac.za DESCRIPTION:Speakers: Jaco Badenhorst (CSIR Next Generation Enterprises an d Institutions Cluster)\nSpeech-enabled human-machine interfaces are becom ing increasingly prevalent. There is\, however\, a disparity between the l anguages these systems are available in and the languages spoken in South Africa. This is because a tremendous effort is required to collect and ref ine large speech corpora. For the majority of South Africa’s languages\, very little speech data is available. This creates a challenge due to the “data hungriness” of automatic speech analysis techniques\, specifica lly those which yield the best performance such as deep learning. Once suf ficient initial resources exist\, automatic data harvesting strategies can be used to further increase the amount of available data in under-resourc ed languages. The aim of the work we report on is to improve existing ASR systems for the 11 official languages of the country. These systems are cu rrently based on the recordings of the NCHLT (National Centre for Human La nguage Technology) corpus\, which consists of an estimated 55 hours of spe ech obtained from 200 speakers for each language. During the NCHLT project additional speech data was collected but not released. To determine wheth er this additional data is useful\, acoustic evaluation of the audio is re quired to detect both low signal-to-noise ratios (SNRs) and transcription errors. The decoding process with state-of-the-art Deep Neural Network-bas ed models is more than an order of magnitude slower than real time on CHPC ’s Lengau Dell cluster’s processors. It requires more than 10 CPU hour s to process one hour of audio data. We therefore used CHPC to run the job s required for processing one language in parallel. This allowed for appro ximately 200 hours of auxiliary data to be processed and used less than 50 GB of memory. The harvested data from this process was subsequently used to train factorized time-delay neural networks (TDNN-F). This model archit ecture was recently shown to yield good results in resource constrained sc enarios (less than 100 hours of speech). These models significantly reduce d phone error rates for the 11 languages. Each TDNN-F model trained in abo ut 8-12 hours on CHPC's Lengau GPU cluster.\n\nhttps://events.chpc.ac.za/e vent/47/contributions/1065/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1065/ END:VEVENT BEGIN:VEVENT SUMMARY:Finite volume modelling of dense granular flow in rotary kilns DTSTART;VALUE=DATE-TIME:20191204T123000Z DTEND;VALUE=DATE-TIME:20191204T125000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1066@events.chpc.ac.za DESCRIPTION:Speakers: Alfred Bogaers (Ex Mente Technologies)\nGranular mat erials are used in several industrial applications. One example of such an industrial application is rotary kilns\, often used for drying\, pre-heat ing and the reduction of a moving\, high-temperature granular bed. The gra nular flow in these reactors have an important influence on capacity\, pro duct quality\, and economic feasibility. Rotary kilns\, in the pyrometallu rgical industry\, often have diameters up to 6m\, with lengths in excess o f 80m\, and operating at temperatures of 1000 to 1400°C. Because of the s ize of these kilns\, modelling the granular flow using the discrete elemen t method (DEM) would result in excessively high computational costs. In th is work\, we therefore made use of a continuum approach to describe the gr anular flow. \n\nWe adopted the μ(I) dense granular flow model proposed b y de Cruz et al. (2005) and later extended by Jop et al. (2006). This mode l is a rate-dependent\, phenomenological description of dense granular mat erials and can be characterised as an elsto-viscoplastic material descript ion with a frictional yield criteria. The flow model approximates an effec tive friction coefficient through a relationship between plastic flow stra in rates and a confinement time scale to account for the internal\, inter- particulate motion. \n\nWe implemented the material model into OpenFOAM\, an open source\, finite volume (FV) based\, partial differential equation toolkit. The volume of fluid (VoF) method was used to capture the discrete granular-fluid interface\, enabling the simulation of large granular bed deformations. The numerical scheme was stabilised by using pressure and vi scosity regularisation\, along with a semi-implicit coupling between the i nternal pressure and velocity fields.\n\nIn our project we were faced with serious technical and computational challenges involving combustion\, hea t transfer\, fluid flow\, high-temperature chemistry\, and the movement of a large granular bed. Our FV approach enabled us to make valuable computa tional modelling and simulation contributions to the development of a new high-temperature process technology.\n\nhttps://events.chpc.ac.za/event/47 /contributions/1066/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1066/ END:VEVENT BEGIN:VEVENT SUMMARY:KEYNOTE 4: Scientific Applications on Heterogeneous Architectures – Data Analytics and the Intersection of HPC and Edge Computing DTSTART;VALUE=DATE-TIME:20191203T074500Z DTEND;VALUE=DATE-TIME:20191203T083000Z DTSTAMP;VALUE=DATE-TIME:20260413T043753Z UID:indico-contribution-47-1076@events.chpc.ac.za DESCRIPTION:Speakers: Michela Taufer (University of Tennessee Knoxville)\n This talk discusses two emerging trends in computing (i.e.\, the convergen ce of data generation and analytics\, and the emergence of edge computing) and how these trends can impact heterogeneous applications. Next-generati on supercomputers\, with their extremely heterogeneous resources and dram atically higher performance than current systems\, will generate more data than we need or\, even\, can handle. At the same time\, more and more dat a is generated at the “edge\,” requiring computing and storage to move closer and closer to data sources. The coordination of data generation an d analysis across the spectrum of heterogeneous systems including supercom puters\, cloud computing\, and edge computing adds additional layers of he terogeneity to applications’ workflows. More importantly\, the coordinat ion can neither rely on manual\, centralized approaches as it is predomin ately done today in HPC nor exclusively be delegated to be just a problem for commercial Clouds. This talk presents case studies of heterogenous app lications in precision medicine and precision farming that expand scientis t workflows beyond the supercomputing center and shed our reliance on larg e-scale simulations exclusively\, for the sake of scientific discovery.\n\ nhttps://events.chpc.ac.za/event/47/contributions/1076/ LOCATION:Birchwood URL:https://events.chpc.ac.za/event/47/contributions/1076/ END:VEVENT END:VCALENDAR