Hi,

We are currently busy with benchmark testing on a VNX 8000 using GPFS for large block sequential workloads.  We are comparing with a similar test done on a VNX 7500. Our results are surprising to say the least as our write, read and read/write throughput on the VNX 8000 all deliver around 40Gbps whereas on the VNX 7500 that we need to beat there were much higher fluctuations between the test results. Please see the summarized results below. What we need to achieve is to level out the performance on the Read/Write test to more closely match the VNX 7500 results.  I have a suspicion that the SSD cache (11x 200GB SSD) is the bottle neck and is in effect throttling the array while being heavily read biased. I would appreciate any comments on this.

The test was done with 8 GPFS Servers first writing 100 concurrent streams down to the array, then 100 Reads, and then at a 70/30 read/write split.

100% write tests: Pass

Randburg: 18 Gbit/s

Samrand: 38 Gbit/s

100% read tests: Fail

Randburg: 48 Gbit/s

Samrand: 40 Gbit/s

70% read - 30% write tests: Fail

Randburg Read: 16.2 Gbit/s

Randburg Write: 13.7 Gbit/s

Randburg Read/Write: 30 Gbit/s

Samrand Read: 32 Gbit/s

Samrand Write: 5.5 Gbit/s

Samrand Read/Write: 37.5 Gbit/s