A Real World Application
- MPI Cluster: Scheduled Transfers on GSN
- Great Performance (From SGI Built Clusters)
- 700 MB/sec single process ..... FAST DATA MOVEMENT!
- % CPU utilization for ST stack and interrupts
- 1.5 million packets per second (I.e. repetition rate of 0.67 usec, 64 byte user payload)
- 4 usec user-process to user process latency
- OS Bypass with no security compromises
- Building Block for 3 Teraflop Program
-
- ST Provides intermedia (GSN, GigE, ATM, HIPPI) interoperability unlike VIA