Principal Investigators: Scott Hemmert, lead, Sandia National Laboratories; Rob Ross, Argonne National Laboratory; Abhinav Bhatele, Lawrence Livermore National Laboratory (LLNL); Nikhil Jain, LLNL; Scott Pakin, Los Alamos National Laboratory
The Interconnect and System Simulation Working Group is responsible to provide ECP teams with networking and full-system scaling expertise and will perform two types of studies: (1) partnering with Facilities, ECP-Application Development/Software Technology, and hardware vendors to study the effects on application performance of system scale and network hardware performance (this will likely include network interfaces as well as switches/routers and associated fabrics and network topology) and (2) provide guidance concerning bottlenecks on exascale-class designs or prototypes.
The basis for these studies are the world leading simulation tools and modeling capabilities developed by several US Department of Energy (DOE) laboratories. Many of these tools and approaches have been exercised in previous Fast and DesignForward DOE programs, delivering vendors accurate and scalable application performance projection capabilities. Application communication traces and/or models (“motifs,” “skeletons” etc.) will be shared with ECP PathForward vendors as requested.