HP comes clean in latest wave of storage benchmarks
The Storage Performance Council has posted another wave of benchmarks, including new results for hardware from IBM Corp and 3Pardata Inc, and a re-test by Hewlett Packard Co in response to criticism of what was diplomatically called the "real world applicability"of its previous results.
While HP set an example for others - namely LSI Corp - to follow by coming clean over a configuration issue, IBM applied the SPC-1 benchmark to the latest version of its Shark ESS storage array, and start-up 3Pardata Inc tested an entry-level configuration of its storage device.
SPC-1 is designed to measure throughput in a simulated OLTP environment. HP's engineers took the company's Enterprise Virtual Array back into the laboratory to re-test it because the first benchmark was completed with write cache mirroring switched off. This is a feature which slows down performance by a fair margin, but which most customers leave switched on to provide improved data protection.
In the first test, when write cache mirroring was switched off, the EVA scored an SPC-1 IOPS figure of 24,005. In the second test when it was switched on, that IOPS number dropped by 16% to just 20,096. The EVA's average response time when handling 10% of the SPC's full benchmark load also slipped, slowing from 2.29ms to 2.36ms.
David Scott, CEO of 3Par, was among those that criticized the first set of HP results as being less than representative. Yesterday he said: "It's admirable of HP to put these new results on the table. Now, the pressure is going to be on LSI to do the same," he said. LSI switched off write cache mirroring when it tested its FastT arrays earlier this year.
Write cache mirroring involves writing of data to two rather than one solid-state caches within a storage array. Without it, data is exposed to a single point of failure. The SPC allows it to be switched off during a benchmark test because it is a supported configuration which customers are not discouraged from using. LSI, which estimates that at least 25% of its customers switch it off in order to improve performance, said yesterday that it has not decided yet whether or not to follow HP's lead.
IBM's test of its latest model 800 version of the Shark ESS returned an SPC-1 IOPS figure of 22,999, up heavily on the 8,009 IOPs recorded for the smaller and much cheaper Shark F20 tested earlier in the year.
Privately-held 3Par tested a medium-sized configuration of its InServ S800 storage server in order to demonstrate what it said is the unmatched scalability of its storage array. In the first test of an eight-controller version of the S800 earlier this year, 3Par scored an SPC-1 IOPS figure of 47,001. In the company' latest test, a smaller two-controller version of the same machine scored an IOPS figure of 12,906. The purpose of the second test was to map out a lower end for what 3Par claims is a massive performance and capacity range that can be provided by just one device undergoing non-disruptive upgrades.