Performance of MPICH-MX 1.2.7..1 over MX-10G
Uniprocessor case: one process per node,
two bonded NICs per node

Performance measurements are presented here for MPICH-MX over MX-10G for the Point to Point Communication tests in the Intel MPI Benchmark (IMB) Suite, Version 2.3.

• Point to Point Communication
  • IMB PingPong
  • IMB PingPing
  • IMB Sendrecv
  • IMB Exchange

We will provide performance data for the Collective Communication tests soon. Each benchmark is run with varying message lengths, and timings are averaged over multiple samples.

The environment for these tests consists of two quad-processor 3.2 GHz Supermicro X7DB8 (Intel Woodcrest) machines with two 10G-PCIE-8A-C NICs in each machine. The two machines were connected point-to-point (switchless). Each machine has 8 GB of memory and was running Linux 2.6.17.11, MX-10G 1.2.0h, and MPICH-MX 1.2.7..1. The Intel MPI Benchmark was compiled with gcc 4.0 with -O.

Notes:

• The data rates (bandwidths) reported by the Intel MPI Benchmarks are in units of mebibytes (2²⁰ Bytes) per second, a peculiarity that the IMB inherited from its predecessor, the Pallas MPI benchmark. For the following performance graphs, the data rate (bandwidth) was converted to standard MegaByte-per-second (MB/s) units.
• Only one process per node and one NIC per node were used in the timings, so there was no intranode communication through shared memory.
• This NIC-bonding mode of operation of MX-10G requires a special MX-10G patch.

Point to Point Communication

Point to point communication performance is measured between two processes. Latency is measured in microseconds (µs, shown as us in the graphs), and bandwidth is measured in MB/s. The latency scale is logarithmic and the bandwidth scale is linear.

IMB PingPong

PingPong is the classical pattern used for measuring startup (latency) and throughput (bandwidth) of a single message sent between two processes.

IMB PingPing

As PingPong, PingPing measures the startup and throughput of a single message sent between two processes, with the crucial difference that messages are obstructed by oncoming messages. For this test, two processes communicate (MPI_Isend/MPI_Recv/MPI_Wait) with each other, with the MPI_Isend's issued concurrently.

IMB Sendrecv

Based on MPI_Sendrecv(), the processes form a periodic communication chain. Each process sends to the right and receives from the left neighbor in the chain. The turnover count is 2 messages sample (1 in, 1 out) for each process.

For 2 processes, Sendrecv will report the bi-directional bandwidth of the system, as obtained by the (optimized) MPI_Sendrecv function. The results below are for 2 processes.

IMB Exchange

Exchange is a communications pattern that often occurs in grid splitting algorithms (boundary exchanges). The group of processes is seen as a periodic chain, and each process exchanges data with both left and right neighbor in the chain.

The turnover count is 4 messages per sample (2 in, 2 out) for each process. The results below are for 2 processes.

Last updated: 27 January 2007