As long as storage systems have been around, their users and managers have always wanted to know how they’d perform under approximate test conditions. If the system is going to be used for email, how does it do with a simulated workload? There are several simulated workloads that are representative of common workloads; database performance, fileserver, and so on. But before we log any results, we first have to run the workload until performance stops rising and falling over time. Once that’s done, we can test the server profile at QD1 through QD128.
The fileserver profile is an 80% read 20% write mix. It’s made up of blocksizes from 512 to 64K, each making up a different percentage of the access pattern. The pattern is: 512 bytes=10%, 1k=5%,2k=5%, 4k=60%, 8k=2%, 16k=4%, 32k=4%, 64k=10%. The Optimus hits over 27,500 IOPS at a QD of 64. From the previous tests, we know that reads scale more with queue depth, and the fileserver pattern is 80% reads.
The database profile is just 8K transfers, but 67% of the operations are reads. If we go back to the SNIA IOPS testing, the 65/35 8K performance is extremely close to what we see here. We get to see the scaling with queue depth here though, and that’s important too.
Traditionally, the email server profile would be used. Lately, we’ve been experimenting with a workload t0 simulate Microsoft’s Exchange Email server. It breaks down 62% read and 38% write.