Sign in with
Sign up | Sign in

Enterprise SSDs: How to Prove Real World Performance

Enterprise SSDs: How to Prove Real World Performance
By
Brought to you by What's this

In our previous article about enterprise SSDs, we touched on the need for drives to perform not only at high throughput levels but consistently in data center settings. A drive that spikes up to stratospheric performance levels for a few seconds under ideal conditions but collapses when tasked with real world enterprise data for days on end doesn’t do an organization any good. Spec sheets only tend to mislead because they give no indication about performance over time or the nature of workload used to obtain those throughput numbers. Certainly, speed matters. This is why SSDs continue to gain popularity in data center environments. But speed must be qualified and examined within the appropriate application context. Otherwise, we might as well be talking about land speed statistics in a boat race

This remains a critical concern for data centers. An IT department must have a fairly accurate idea of how a drive will perform under fire before committing to deployment. Benchmark tests have always been the answer to this need, but which tests can or should be trusted? There’s a lot more to this question than meets the eye.

Why Your Benchmarks Are Probably Wrong

To be fair, there is no such thing as a perfect synthetic benchmark. At best, one can hope for a close approximation of real world performance from a canned test, but consider this common scenario: A small enterprise needs SSDs to fuel its new e-commerce server system. This system will run around the clock, and it will be subject to periodic spikes in traffic. Much of the server’s load is random data, but occasional batch jobs of large sequential data can also flow into the drive.

How does an average business determine whether a prospective SSD is suitable for this work environment? More often than not, it will start with media reviews that tested the drive with tools such as Futuremark’s PCMark and Iometer. PCMark is better than many synthetics in that is uses actual program executables with real world-type data. However, PCMark’s application set includes programs such as Windows Defender and Windows Media Player, which are hardly representative of the task types performed by an e-commerce server.