Download
high performance computing at ttu n.
Skip this Video
Loading SlideShow in 5 Seconds..
High Performance Computing at TTU PowerPoint Presentation
Download Presentation
High Performance Computing at TTU

High Performance Computing at TTU

113 Views Download Presentation
Download Presentation

High Performance Computing at TTU

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. High Performance Computing at TTU Philip Smith Senior Director HPCC TTU

  2. Brief History • Founded in 1999 • First cluster in 2001 • TechGrid in 2002 • Statewide grid project in 2004, funded in 2005 • Large cluster with IB in 2005

  3. Major Users • Molecular Dynamics (MD) • Chemistry, Physics, Chemical Engineering • Quantum Dynamics • Numerical Weather Prediction • Mechanics (FEM simulations) • Energy exploration (reservoir modeling) • Beginning to see bioinformatics applications

  4. Science at Texas Tech University

  5. Expansion in 2008 • $1.8 M project to add a new computer room • Completion date 7/15/08 • Move in 8/15/08 • Double or triple our compute capacity in September 2008 with expansion room for several years.

  6. Current Resources Hrothgar Shared 648 cores ~ 2 GB per core Community Cluster 188 cores ~ 2 GB per core 12TB shared storage SDR Infiniband + GigE management network

  7. Current Resources (cont) • Antaeus (shared grid resource) • 264 cores ~2GB/core • 6 TB shared storage, 44 TB dedicated storage • GigE network only • TechGrid • 850 lab machines • Distributed between 5+ sites on campus • 10/100 network

  8. Thank you • Questions?

  9. TechGrid • Cycle scavenging grid • Computing labs in BA, Math and library • ~650 machines • Avaki → Condor in 2007

  10. Limitations on TechGrid • Windows binaries • Most machines only available 11pm-7am • Loosely coupled machines: no mpi • 1GB or less of memory

  11. General porting issues • Getting it to run: • Does it have a Window's Binary? • Does it require modification to compile? • Production usage: • Data foot print (input/output) • Run time • Numbers of iterations • Interaction between iterations

  12. Cluster job mix • 65% MD, 25% QD and 10% other • Currently about 56% MD with balance QD • >90% parallel

  13. Campus grid job mix <10% utilized Open source applications currently used Venus R SAS grid