SCI 2013Jul30 Latest SPC-1 performance results

In $/IOPS, 3PAR, DS8000, DS8800, DS8870, Hitachi Vantara, Hitachi VSP, HPE, Huawei, HUS VM, IBM, IOPS, LRT, OceanStor Dorado2100 G2, SPC, SPC-1, StoreServ 7400, XIV, XIV Gen3 by AdministratorLeave a Comment

We return now to Storage Performance Council (SPC) results*.  There have been five new SPC-1submissions, the HDS HUS VM, HDS VSP with flash, HP 3PAR StoreServ 7400, Huawei OceanStor Dorado2100 G2, and IBM XIV Gen3. There were a few new SPC-1/C submissions from Seagate and no new SPC-2 submissions over this past quarter. It turns out just about every one of our top 10 charts for SPC-1 have new results.

SPC-1*results

We start our discussion with the Top 10 Least Response Time (LRT) chart for SPC-1.

Figure 1 SPC-1* Top 10 LRTBar chart with SPC-1 Least Response Time, Top 10 SPC-1 LRT

Lower is better on the Top 10 LRT chart.  All of these systems are SSD only or SSD-disk hybrids. There were four new systems in this top 10, including the HDS VSP with flash at #4 with an LRT of 0.28msec, HP 3PAR 7400 at #5 with an LRT of 0.33msec, the Huawei OceanStor Dorado G2 at #8 with an LRT of 0.38 msec. and IBM XIV at #9 with an LRT of 0.41msec. The XIV is noteworthy here as the only system in our top ten to combine SSDs and 2TB disk drives.  The other notable item here is that of the new systems they differ by only 121 μsec in LRT.

As discussed last time, the LRT chart has become dominated by SSD or DRAM system results.  It seems that even at the 10% system activity level (where SPC-1 measures LRT), the workload introduces some amount of backend overhead that is difficult to avoid – no doubt, due to some amount of read miss and/or destage activity.

Next we turn to top 10 IOPS in Figure 2.

Top SPC-1 IO operations per second, SPC-1 Top 10 IOPSFigure 2 SPC-1 Top 10 IOPS

There were only twe new result in our IOPS Top 10, which were the HDS VSP with flash that hit over 600K IOPS and the Huawei OceanStor Dorado2100 G2 which generated ~400K IOPS. The HDS VSP had 24 Flash memory modules and the Dorado2100 had 50 SSDs for their backend storage.

There are still a few disk-only storage systems on this chart (#5-7 and #10). However, most of these top 10 IOPS disk-only systems had ~2000 disk drives to get to this level with the lone exception being the IBM DS8870 which had only ~1500 disks.  .

Next we turn to our Top 10 $/IOPS in Figure 3.

Figure 3 Top 10 SPC-1 $/IOPSTop SPC-1 $ per IO operations per second, SPC-1 Top 10 $/IOPS, SPC-1 price performance

Haven’t examined $/IOPS™ in a while as we prefer our own IOPS/$/GB ranking but this is one of the key reported SPC-1 metrics so we report on it occasionally.  Both the Huawei OceanStor Dorado2100 G2 at #2 and HP 3PAR StorServ 7400 at #3 are new to this chart. SSD only systems seem to dominate this chart as they generate a lot of IOPS and due to the small capacity in these submissions, cost relatively little in comparison to an disk-only array. Indeed, all of the top 10 in $/IOPS are SSD only storage systems.

Next we turn to one of our perennial favorite charts IOPS vs. LRT bubble chart in Figure 4.

Comparing Cost/GB, IOPS performance and Response time (LRT) in a bubble chartFigure 4 Top 10 SPC-1 IOPS vs. LRT bubble chart

Disk-only subsystems on this chart are in pink and all-flash or flash-disk hybrids are in light green. But they all cost less than $100/GB. We decided to differentiate the disk-only from the hybrid and SSD-only storage systems so we could see how they influence LRT. At the 1msec LRT level there are a few disk-only storage systems that can compete well with the all-flash and flash-disk hybrid storage systems, especially at the higher IOPS. But as the disk-only systems generally have over 1000 disk drives to get to the higher IOPS their cost is substantially higher.

Some of the other all-flash/all-dram performers were too expensive to make this chart, which is also true of some of the disk-only storage systems. Nonetheless, if you need an LRT under 1 msec. one would probably need to go to an all-flash, all-DRAM or flash-disk hybrid system to get it.

Significance

It’s good to see new SPC-1 data in our top 10 charts.  This new batch of SPC-1 submissions had a good showing over all for both the all-flash and flash-disk hyrid systems. There was only one all-disk submission, that being the HDS HUS VM and only one hybrid system, the IBM XIV.

Flash-only, flash-disk hybrids and DRAM-only systems are dominating the LRT charts and it’s almost time to split out the disk-only systems into their own special LRT chart. However, latency is starting to become important again (was it ever not important?).  Nonetheless, there needs to be some way to show off the better disk-only storage systems as they come up.  For instance, the HDS HUS VM had an LRT of 1.2 msec. which would rank #4 in a top ten chart for disk only storage LRTs but shows up at #25 in a Top 30 LRT ranking.

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 of all these charts and a set of new ChampionsCharts™ for SAN storage in our recently updated (February 2019) SAN Storage Buying Guide available from our website. Also more performance details for some of these charts are available in our newly revised (December 2019) SAN-NAS Buying Guide, available for purchase from our website.

~~~~

[This performance dispatch was originally sent out to our newsletter subscribers in July of 2013.  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. Dispatches are posted to our website at least a quarter or more after they are sent to our subscribers, so if you are interested in current results please consider signing up for our newsletter.]  

Silverton Consulting, Inc. is a Storage, Strategy & Systems consulting services company, based in the USA offering products and services to the data storage community.


* All results from www.storageperformance.org as of 30Jul2013

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.