BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:Programmable Storage Systems DTSTART;VALUE=DATE-TIME:20171204T092500Z DTEND;VALUE=DATE-TIME:20171204T095000Z DTSTAMP;VALUE=DATE-TIME:20260608T121336Z UID:indico-contribution-61-22@events.chpc.ac.za DESCRIPTION:Speakers: Carlos Maltzahn (University of California\, Santa Cr uz)\nStorage systems need to support high-performance for special-purpose data processing applications that run on an evolving storage device techno logy landscape. This puts tremendous pressure on storage systems to suppor t rapid change both in terms of their interfaces and their performance. Bu t adapting storage systems can be difficult because unprincipled changes m ight jeopardize years of code-hardening and performance optimization effor ts that were necessary for users to entrust their data to the storage syst em. We introduce the programmable storage approach\, which exposes interna l services and abstractions of the storage stack as building blocks for hi gher-level services. We also build a prototype to explore how existing abs tractions of common storage system services can be leveraged to adapt to t he needs of new data processing systems and the increasing variety of stor age devices. We illustrate the advantages and challenges of this approach by composing existing internal abstractions into two new higher-level serv ices: a file system metadata load balancer and a high-performance distribu ted shared-log. The evaluation demonstrates that our services inherit desi rable qualities of the back-end storage system\, including the ability to balance load\, efficiently propagate service metadata\, recover from failu re\, and navigate trade-offs between latency and throughput using leases.\ n\nhttps://events.chpc.ac.za/event/6/contributions/22/ LOCATION:Velmoré Hotel Estate Rendezvous URL:https://events.chpc.ac.za/event/6/contributions/22/ END:VEVENT BEGIN:VEVENT SUMMARY:EMPRESS--Extensible Metadata PRovider for Extreme-scale Scientific Simulations DTSTART;VALUE=DATE-TIME:20171204T095000Z DTEND;VALUE=DATE-TIME:20171204T101500Z DTSTAMP;VALUE=DATE-TIME:20260608T121336Z UID:indico-contribution-61-23@events.chpc.ac.za DESCRIPTION:Speakers: Jay Lofstead (Sandia National Laboratories)\nSignifi cant challenges exist in the efficient retrieval of data from extreme-scal e simulations. An important and evolving method of addressing these challe nges is application-level metadata management. Historically\, HDF5 and Net CDF have eased data retrieval by offering rudimentary attribute capabiliti es that provide basic metadata. ADIOS simplified data retrieval by utilizi ng metadata for each process' data. EMPRESS provides a simple example of t he next step in this evolution by integrating per-process metadata with th e storage system itself\, making it more broadly useful than single file o r application formats. Additionally\, it allows for more robust and custom izable metadata.\n\nhttps://events.chpc.ac.za/event/6/contributions/23/ LOCATION:Velmoré Hotel Estate Rendezvous URL:https://events.chpc.ac.za/event/6/contributions/23/ END:VEVENT BEGIN:VEVENT SUMMARY:Scientific Modeling of Storage System Reliability DTSTART;VALUE=DATE-TIME:20171204T090000Z DTEND;VALUE=DATE-TIME:20171204T092500Z DTSTAMP;VALUE=DATE-TIME:20260608T121336Z UID:indico-contribution-61-25@events.chpc.ac.za DESCRIPTION:Speakers: Matthew Curry (Sandia National Laboratories)\nWhen d esigning large scale storage systems\, failure is always a serious concern that demands constant attention. However\, the ability for system designe rs to objectively evaluate their risk of data loss for a given storage sys tem is minimal. Instead\, they must enumerate possible failure modes\, est imate their relative probability\, identify possible mitigations\, and dec ide whether the expense is worthwhile. This process relies on folk wisdom\ , rules of thumb\, and anecdotal experience. For systems that grow larger and more complex year-by-year\, this methodology is too imprecise to guara ntee safety while ensuring efficiency.\n\nThis talk will detail some of th e progress in the SIMS^2 project\, a collaboration between Sandia National Laboratories\, University of Wisconsin-Madison\, and Los Alamos National Laboratory charged with increasing the science and rigor behind evaluating system designs. It will cover some of the pitfalls of current methods of evaluating systems\, methods for determining complex behavior of aggregate d components\, evaluation of different types of failure modes\, and some i nteresting inflection points for different system designs.\n\nhttps://even ts.chpc.ac.za/event/6/contributions/25/ LOCATION:Velmoré Hotel Estate Rendezvous URL:https://events.chpc.ac.za/event/6/contributions/25/ END:VEVENT BEGIN:VEVENT SUMMARY:HPC Storage – the two distinct challenges DTSTART;VALUE=DATE-TIME:20171204T101500Z DTEND;VALUE=DATE-TIME:20171204T104000Z DTSTAMP;VALUE=DATE-TIME:20260608T121336Z UID:indico-contribution-61-107@events.chpc.ac.za DESCRIPTION:Speakers: Wolfgang Mertz (CTO HPC Dell EMC Isilon)\nTraditiona lly when architecting the storage for an HPC environment the focus is on p erformance of the file system. There is another challenge though\, as one has to manage all the files that are used for input or created as output\ , typically over a long period of time. There is even value in this data that can be mined via modern analytics methods (aka Hadoop) or used for AI /ML. This session will outline why it is essential to use different solut ions to address “the performance challenge” and “the persistent data challenge”.\n \nBiography: Wolfgang holds a PhD on computer aided engin eering. In his more than 25 years in the IT industry he held various tech nical and strategic positions. He focuses on understanding the verticals ’ specific processes\, needs and applications\, which he then links to t he IT product’s features and capabilities in order to achieve the best r esults for the customers. In his current role as CTO for EMEA in Dell EMC ’s Unstructured Storage Team he specializes mainly HPC and related verti cals. Wolfgang also follows the immense transformations IT is currently u ndergoing\; this especially includes the new capabilities data lakes\, big data analytics and AI/ML offer.\n\nhttps://events.chpc.ac.za/event/6/cont ributions/107/ LOCATION:Velmoré Hotel Estate Rendezvous URL:https://events.chpc.ac.za/event/6/contributions/107/ END:VEVENT END:VCALENDAR