SCI 20Nov2015 Latest SPC performance analysis

In $/IOPS, 3PAR, ETERNUS DX100 S3, FlashSystem, FlashSystem 820, FlashSystem 840, Fujitsu, HPE, Huawei, IBM, Kamanario K2-D (DRAM), Kaminario, LRT, OceanStor Dorado2100 G2, SC4020, SPC, SPC-1, SSD storage, StoreServ 7400 by Administrator2 Comments

This Storage Intelligence (StorInt™) dispatch covers Storage Performance Council (SPC) results[1]. There have been four new SPC-1 and no new SPC-2 submissions since our last report in August [although one of the SPC-1/E reports was not new, it was missed previously, ed.]. The new SPC-1 results are all-flash solutions and include Dell SC4020, Fujitsu ETERNUS DX100 S3 and IBM FlashSystems 820 [missed previously] and 840. Three of our top ten charts have changed and we present them below.

SPC-1 results

We begin our discussion with LRT™ (least response time) top ten results shown in Figure 1.

Top 10 bar chart with SPC-1  Least Response Time LRT resultsFigure 1 Top 10 SPC-1 LRT™ results

First as discussed in previous reports, all of our top ten LRT systems have a majority of their ASU capacity in SSDs or Flash storage. The new submissions from IBM, [previously overlooked] FlashSystem 820 and 840 rank as second and fifth best in LRT. Recall that LRT is measured at the host and at 10% of system IO load. The difference between the IBM FlashSystem 820 and 840 is measured at 50 μsec. The top 5 best response times are all from current or previous (pre- and post-acquisition) generations of IBM FlashSystem solutions.

Next, we turn to $/IOPS™, a SPC cost of performance reported metric in Figure 2.

Figure 2 Top 10 SPC-1  $/IOPS™Top 10 SPC-1 Price Performance $/io operation per second results

In Figure 2 we see both the Dell SC4020 and the Fujitsu DX100 S3 came in fourth and tenth place respectively. Similar to LRT results above, the $/IOPS is dominated by all-flash systems. This metric divides the solution price by its achieved IOPS and as all-flash systems generate substantial IOPS regardless of capacity, smaller all SSD/Flash systems rank well here. Both the Dell and Fujitsu systems had relatively little storage, with both just over a TB of ASU capacity each, that keeps costs down. However as all-flash systems they were able to generate over 105K and 55K IOPS respectively, which helped them place well here.

We don’t report on this metric often because it doesn’t consider capacity at all. We prefer our IOPS/$/GB, which we feel better, represents the economics of storage system performance. We show our top 10 IOPS/$/GB in Figure 3.

Top 10 SPC-1 (SCI) price performance results IO operations per second per capacity costFigure 3 Top 10 SPC-1 IOPS/$/GB

In Figure 3 we can see the new IBM FlashSystem 840 coming in sixth place. Note, our IOPS/$/GB also only ranks two all-flash systems (Kamanario and IBM FlashSystem) solutions in the top 10. The number one system (Huawei) had the third highest IOPS but was a hybrid system and the third through fifth ranked systems (HP 3PAR and Huawei OceanStors) were disk-only storage systems.

We believe due to this reduced bias to all-flash systems, the IOPS/$/GB metric provides a better view of the economics of storage performance than the SPC-1 reported $/IOPS metric. Although, a difference of opinion still exists on this.

Significance

Sorry we missed the earlier report on the IBM FlashSystem 820. Our only excuse is that it was an SPC-1/E report but other SPC-1/E results are included here and it should have been as well. The SPC-1/E report focuses on energy consumption, at various IOPS levels. We haven’t reported on energy consumption results in a while mainly because there are so few of them (seven) and because of their reduced energy consumption, all-flash dominate this category as well.

It seems everyone is fielding all-flash arrays these days. The mid-range to lower, Dell SC4020 and Fujitsu DX100 S3 storage systems have some pretty decent performance when configured with all-flash storage. That these submissions only had ASU capacities in the TB range are probably indicative of their target market segments. Our only question is whether an all-flash shared storage system is a reasonable solution for these market segments or whether a modest quantity of PCIe server side flash with appropriate caching software wouldn’t be a better fit.

As always, suggestions on how to improve any of our performance analyses are welcomed. Additionally, if you are interested in more block storage performance information, we now provide a fuller (top 30 results) version of some of our performance charts and a set of new, SCI derived, OLTP, Throughput and Email ChampionsCharts™ for enterprise, all-flash, mid-range and SMB SAN storage, in our updated (November, 2014), SAN Storage Buying Guide available for purchase from our website[2].

[Also we offer more block storage performance information plus our OLTP, Email and Throughput ChampionsCharts™ charts in our recently updated (May 2019) SAN Storage Buying Guide, or for more information on some select ESRP performance results please see our recently updated (May 2019) SAN-NAS Storage Buying Guide, both of which are available for purchase on our website.]

[This performance dispatch was originally sent out to our newsletter subscribers in November of 2014.  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 U.S.-based Storage, Strategy & Systems consulting firm offering products and services to the data storage community

[1] All SPC results available from http://www.storageperformance.org/home/ as of 20Nov2014

[2] Available at http://silvertonconsulting.com/cms1/product/san-storage-buying-guide/

Comments

Leave a Reply

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