Earth Simulator Rapid Response Meeting

1
Earth Simulator Rapid Response Meeting

• Buddy Bland
• May 15, 2002

2
US Capabilities 2003-2004
3
Five Year Vision for US HPC

• Requires strategic investments to improve
  competitiveness of HPC resources
• System Balance
• Memory bandwidth and latency
• System interconnects
• I/O capabilities
• Focused use of resources
• Experimental architectures
• HPC System Software

4
Experimental Architectures

• Required in a 5 year strategy to achieve
  PetaFLOP/s
• Blue Gene/D generalizes the QCDOC architecture
• Dramatic price/performance
• IBM PPC processor with enhanced floating point
• Cray SV2 offers first scalable vector system
• Price is high for peak FLOPS but may be
  competitive for delivered performance

5
HPC System Software Needs

• Fault Tolerance for highly scalable systems
• Scalable, Standardized Management tools for
  efficiently running our large computing centers
• Robust Parallel File Systems to replace todays
  fragile ones
• Distributed, Dynamic Data Analysis and Storage
• High-speed, wide-area file transfer
• Analysis tools at the centers where the data is
  located
• HPSS needs to add grid services and be seamless
  throughout DOE
• Distributed visualization
• Unified authentication system across DOE centers
• migrate to PKI authentication using a common tool
  set by 2004
• Secure remote access
• SSH and and variants do not fully interoperate
  and are complex