Runtime Systems and Tools: BlueWaters - Sustained Petascale Computing
Prof Kale and PPL participated in developing the successful proposal for the NSF's sustained-petaFLOPS "Track 1" system, named Blue Waters, which will be deployed at NCSA. Some of the proposed work that PPL will be involved in includes:
  • Use of PPL's BigSim to allow early application developers to identify performance bottlenecks. BigSim allows one to develop and run (emulate) at-scale applications and test them on the smaller machines available today; It includes a simulation system that uses the traces obtained during simulation, optionally with cycle-accurate simulation data for sequential-execution-blocks, and optionally a detailed network model, to produce performance predictions at multiple resolutions.
  • Develop and deploy Charm++ on Blue Waters, as well as collaborate in development of a virtualized MPI (based on the AMPI model). These will support adaptive overlap of computation and communication as well as automated resource management.
  • Scaling of NAMD to the full-scale machine, in part to ensure that the NSF 100-million atom benchmark performs at the desired level.
In addition, PPL is assisting the project in various ways.
People
Papers/Talks
11-47
2011
[Talk]
Avoiding Hot-Spots on Two-Level Direct Networks [SC 2011]
11-21
2011
[Paper]
Avoiding Hot-Spots on Two-Level Direct Networks [SC 2011]