Panel at Internet2 Spring Meeting, April 2010
?Download as PPTX, PDF?
0 likes?224 views
Scientific applications increasingly rely on large datasets that require high-speed networks for remote collaboration and distributed analysis. Dr. Allen's research in multi-physics simulations, data archiving, and visualization faces challenges due to the complexity of networking multiple sites. There is a need for high-level application services and APIs to integrate networks and enable new science scenarios beyond simply moving files. Demonstrating working prototypes will help address technical details and allow computational scientists to fully leverage network capabilities.
Panel at Internet2 Spring Meeting, April 2010
- 1. Scientific Applications & High Speed NetworksDr. Gabrielle AllenDepartments of Computer Science andPhysics & AstronomyCenter for Computation & TechnologyLouisiana State UniversityNSF 0947825 (EAVIV), 0904015 (Einstein Toolkit)
- 2. Whyare Networks Important for Science?Science & engineering increasing data driven＾data tsumani￣, NSF DataNetCannot store all data, stream processingNon-traditional applications e.g. music, artFundamentally large dataRemote collaboration is crucialNew capabilities neededReally interactive!Exascale: workflow assumes remote analysisMany current scientific use casesPetascale & multicoreissues aheadUse cases crucial for my scientific research
- 3. Multi-physics SimulationsMulti-phase fluids for neutron star cores(nuclear densities, radiation transport, neutrino transport)Modeling Gamma Ray BurstsCosmological spacetimesDifferent models use different algorithms, better suited to particular architecturesGravitational waves traveling through vacuum spacetimeNeed: Co-scheduling, data transfer between components
- 4. Data ArchivingLarge scale resourcesRangerQueen BeeKrakenLocal analysisGroup DataCommunity ArchivePublished DataGroup DataTerabytes of data to be moved (and described).
- 5. Distributed Analysis and VisualizationAndrei will describe: interactivity, large distributed data, co-scheduling, many challenges
- 6. CollaborationEinstein Toolkit Team: multiple sites collaborating in US, Europe, JapanNeed technologies that allow real interaction, sharing whiteboard, easy and quick to set up
- 7. ChallengesComputational scientists more comfortable with clusters/filesystems than networksLimits science! Scaled down resourcesNetworking currently involves many partners, long timescales. International connectivity very hard.High level application-oriented services needed.APIs between services and applicationsCrucialEnd-to-end capabilities (Science done on workstations/laptopsPersistency and production qualityAdditional tools, e.g. advanced reservations, on-demand provisioning.Need to have real working prototypes to be able to get to details ´ not just about moving files, new science scenarios