Ohio State University

HPCC Random Ring | Performance | Network-Based Computing Laboratory

HPCC Random Ring Benchmark

  • Experimental Testbed: Each node of our testbed has two 8-core 2.6 GHz Intel Xeon E5-2680 (Sandy Bridge) processors and 32 GB main memory. The nodes support 16x PCI Express Gen3 interfaces and are equipped with Mellanox ConnectX-3 FDR HCAs with PCI Express Gen3 interfaces. The operating system used was CentOS release 6.3.
  • The Random Ring benchmark is latency benchmark which is part of the popular HPCC Challenge Benchmark Suite.
  • InfiniBand (IB) offers many communication modes of which Reliable Connected (RC) and Unreliable Datagram (UD) are two of the more popular ones. MVAPICH2 uses the RC IB Queue Pair (QP) for all communication by default. As part of our on-going research, we have enabled MVAPICH2 to use a combination (hybrid) of RC and UD QPs based on the communication pattern applications. MV2_HYBRID_ENABLE_THRESHOLD defines the threshold for enabling the hybrid transport. Hybrid mode will be used when the size of the job is greater than or equal to the threshold. Please refer to the following section of the userguide for more information on running MVAPICH2 in hybrid mode.
  • The bar graph depicts the performance improvements that can be obtained for the HPCC Random Ring benchmark by using a hybrid of RC and UD QPs instead of just RC or UD QPs.