SC Conference - Activity Details

Future Networking Schemes in Petascale Clusters and Beyond

Primary Session Leader:
Gilad Shainer  (Mellanox Technologies)

Secondary Session Leaders:
Richard Graham  (Oak Ridge National Laboratory)
Ola Torudbakken  (Sun Microsystems)
Birds-of-a-Feather Session
Wednesday,  12:15PM - 01:15PM
Room Ballroom F
Future HPC systems will span tens-of-thousands of nodes, all connected together via high-speed connectivity solutions to form multi-Petaflop clusters. With the growing size of clusters and CPU cores, not only the traditional demands from the cluster interconnect increase dramatically, but new demands are introduced. The interconnect needs to provide balanced throughput, scalable latency and efficient data handling to address the netwroking requirements of the CPU cores, while maintaining high network utilization. Moreover, the overall number of communication links grows with the size of the cluster and link data errors have become a growing concern for large-scale platforms, as they tend to have an adverse affect on the performance. The session will drive a discussion on the needed communication capabilities, static, multi-static and dynamic routing, congestion control and congestion avoidence, and handling networks errors. The session will also present applications usage cases with the new adaptive routing implementation for InfiniBand 40Gb/s.
   IEEE Computer Society  /  ACM     2 0   Y E A R S   -   U N L E A S H I N G   T H E   P O W E R   O F   H P C