Sign in with
Sign up | Sign in

More Real Results

Why You Need a Reliable Storage Benchmark
By
Brought to you by What's this

The data and graphing then repeats for I/O request throughput distribution, then again for average response time distribution (sustainability). One “problem” with SSD response performance plotted on a graph also made to accommodate hard drives is that you end up with a flat line along the bottom of the graph because SSD response times are so much faster than conventional disk technology, and the graph was scaled to accommodate both technologies. This is why the SPC provides a more granular view into the results by breaking down the millisecond-level responses into a distribution curve.

Next, the SPC report turns to IOPS, again detailing data on a by-minute basis and providing key graphs. Note in the following view how the different types of workloads in each ASU yield markedly different performance results.

Perhaps most importantly, the report then breaks down I/O throughput at six incremental load levels. It’s obviously infeasible for any testing organization to have a methodology for every enterprise storage application, which would be ideal for determining true real world performance. But these load levels provide the best compromise available. IT staff should perform their own usage assessments to correlate against the SPC’s usage levels.

Repeatability tests essentially ensure that the data best represents device performance. For example, you’ll recall that the Pulsar XT.2’s overall reported IOPS was 20,009 (rounded). In reality, the repeatability test showed 20,004 IOPS on the first pass and 20,012 on the second. This is clearly within an extremely tight margin of error and proves that the primary results are reliable.

The data persistence tests entail writing nearly 48 million logical blocks. Because some of these involved overwriting, over 30 million logical blocks could actually be verified, meaning that the system could confirm that the written data and source data perfectly matched. In this case, there were no logical blocks that failed verification.

Finally, the report finishes with a confirmation of the drive’s available shipping date and notes about any anomalies encountered during testing (none in this).