White Papers

12
Figure 7. Sequential read performance (N-1)
Sequential read performance summary:
• Peak read performance of F800 was ~5.6 GB/sec
• Peak read performance of H600 was ~2 GB/sec
• Read performance of H600 did not increase with more than 2 clients
Future tests will utilize more than 64 client compute nodes in an attempt to maximize read throughput on the F800.
Random I/O Performance Tests
Similar to the sequential IOR tests, there are 7 random IOR test cases (1-client, 2-client, 4-client, 8-client, 16-client, 32-client, and 64
client test cases), but the random F800 and H600 tests in Figure 8A used a 4 KB block size and each client writes a 32 GB file. The
random H600 and NSS-7.0-HA tests in Figure 8B used a 4 KB block size and each client writes an 8 GB file. The F800 configuration
was changed to enable better random I/O performance by executing the following OneFS commands affecting endurant cache,
coalescer and multi-writer
6
:
o isi_sysctl_cluster efs.bam.ec.mode=0
o isi nfs settings export modify --setattr-async=1 --commit-async=1
o isi_sysctl_cluster efs.bam.coalescer.multiwriter=0
o isi set -c coal_only /ifs/iotest