We return now to the preeminent block storage benchmark, the Storage Performance Council SPC results*. There have been three new SPC-1 submissions since our last report, one from IBM®, an 8-node SVC 6.2 configuration with 16-Storwize® V7000 systems behind it, one from Huawei Symantec, an all SSD version if their Oceanspace™ Dorado2100 and one from Fujitsu, on their latest generation ETERNUS DX80 S2 storage. There were no new SPC-2 results.
We start our discussion with the top 10 IOPS™ performance metric for SPC-1.
Figure 1 Top 10 SPC-1* IO/Sec Higher is better on the IO/Sec or IOPS results. Here we can see the new IBM SVC 6.2 storage configuration come in as the new #1 at over 520K IO/second. Of course the fact that it had 1950 disk drives might have made a difference here but all the others in the current top 10 either had over 1000 drives or were using SSDs. So disk count is probably not the only advantage here. Eight nodes certainly help, but HP 3PAR and Huawei Symantec’s Oceanspace also had eight, potentially another advantage. However, for whatever reason, previous versions of SVC have done well on the top 10 IOPS chart for some time now, so more likely sophisticated caching, and more total cache (the IBM system used a total of ~450GB of DRAM cache) might be the real secret to doing well here. Next we turn to storage system access latency.
Figure 2 Top 10 Least Response Time (LRT) results Lower is better on the LRT™ chart. Here we can see the impact of all SSD systems and the new Huawei Symantec Dorado2100 system came in at #6 with just over a 0.6msec LRT. The top 7 storage systems (including the NetApp FAS3270A) on the LRT chart either used all SSDs or a FlashCache in their configuration. It’s not until you get to the IBM DS8K systems that one sees any disk only systems and all of these (#8,9&10) are under 1msec a very good response time for non-SSD/non-FlashCache storage. Next we turn to a measure of the cost of performance.
Figure 3 SPC-1 $/IOPS™ Lower is better on the $/IOPS™ chart. We don’t often report on this SPC-1 published metric as we prefer our own GB/$/IOPS but two of the new results show up well on this chart the new Huawei Symantec Oceanspace Dorado2100 and the new Fujitsu ETERNUS DX80 S2 at #2 and 7 respectively. Unclear what the secret is to performing well here but one has to generate relatively high IOPS rates for the price of the storage system. The two Sun J4K storage systems (#6&8) did well on this chart but did not use any RAID protection whatsoever (an unlikely customer deployment). The many all SSD systems here (#1-5, & 10) performed admirably but all had relatively modest amounts of storage driving system cost down and generated serious IOP rates. The two more normal storage systems showing up on this chart, the Fujitsu’s DX80 S2 and older Huawei Symantec S2600 had relatively small amounts of storage, ~10TB and ~3TB respectively which certainly brought system price down and reasonable IOPS performance.
Figure 4 Scatter plot: IOPS vs. number of disk drives Finally, we include our IOPS vs. drive count scatter plot. We like to display this chart to help answer the question as to whether subsystem IOPS performance is driven purely by spindle count. Alas, as shown above, an R**2 of 0.96 clearly indicates that SPC-1 IOPS rate is mainly governed by disk count. A couple of caveats on this chart, we eliminate any SPC-1 result using SSDs, FlashCache or multiple drive capacities so as not to undermine the results. Also this chart only contains results for drives larger than 140GB (active SPC-1 results go all the way back to 36GB disks). The wider variance at higher IOPS rates seems to be informative. Consider the pair of results at ~275K IO/second. I consider the higher performer to be the one using less disk spindles, although it achieved ~270K IOPS vs. its brethren at ~275K IOP which were the HDS USP-V dynamic provisioned vs. an older IBM SVC 4.1 storage. Although, 5K more IOPS can’t be ignored, the fact that the HDS system almost matched it with ~400 less drives indicates that it was the better performer.
It’s good to see some new SPC-1 results especially at the high-end, enterprise class level. With these high level systems the IOPS seems to attain new heights with each new round of submissions. The other nice thing is at least for IOPS rates, disk drive only systems are holding their own in the Top 10. In contrast, the Top 10 LRT chart is now dominated by all SSD or hybrid FlashCache/disk systems, with these systems now holding the top 7 slots. We don’t often discuss the $/IOPS metric because of its dominance by smaller capacity systems but for this report two of the three new results did well. As always, suggestions on how to improve any of our performance analyses are welcomed. Additionally, if you are interested in more block performance details, we now provide a fuller version (top 30 results) of all these charts and more in our recently revised (November 2018) SAN Storage Buying Guide available from our website.
[This performance dispatch was originally sent out to our newsletter subscribers in Febuary of 2012. If you would like to receive this information via email please consider signing up for our free monthly newsletter (see subscription request, above right) and we will send our current issue along with download instructions for this and other reports. Also, if you need an even more, in-depth analysis of SAN storage system features and performance please see our SAN Storage Buying Guide available on our website.]
Silverton Consulting, Inc. is a Storage, Strategy & Systems consulting services company, based in the USA offering products and services to the data storage community