White Papers

11
Figure 6. Sequential write performance (N-1)
Write performance for both the F800 and H600 peaked with the 2-client test case at approximately 1.1GB/sec and didn’t change much
as more compute nodes were added. It appears that file synchronization overhead limits write performance as an increasing number of
client nodes attempt to concurrently write to the same file. In contrast, the N-to-1 read performance generally increases with an
increase in client nodes and peak read performance was approximately 5.6 GB/sec (Figure 7). Future tests will utilize more than 64
client compute nodes in an attempt to maximize N-to-1 write throughput.
Sequential write performance summary:
• Peak write performance of F800 was ~1 GB/sec
• Peak write performance of H600 was ~1.1GB/sec
• Unlike sequential N-to-N write performance, sequential N-to-1 write performance of both H600 and F800 did not increase with
more than 2 clients
Figure 7 shows the IOR read performance for the N-to-1 tests.