SCI’s latest SPC-1 performance report as of Feb’19

In $/IOPS, Huawei, IOPS, LRT, OceanStor, OceanStor 5600 v5, SPC, SPC-1, TTA Jet-speed by AdministratorLeave a Comment

This Storage Intelligence (StorInt™) dispatch covers Storage Performance Council (SPC) results . Over the last quarter there have been 3 new SPC-1 v3 submissions, Huawei OceanStor 5600 v5, Telecommunications Technology Association (TTA) Jet-speed™ HHS3124F/HHS2112F (10 nodes) and TTA Jet-speed HHS3124F/HHS2112F (7 nodes) storage systems. Many of our top 10 SPC-1 charts have changes, a select few of which are shown below. Unfortunately, there were no new SPC-2 submissions since our last report.

SPC-1 results

We begin our discussion with top 10 SPC-1 IOPS™ performance in Figure 1.

Figure 1 Top 10 SPC-1 IOPS

The TTA HHS3124F/HHS2112F (10-nodes) storage system came in as our new #7 with 2.4M IOPS. None of the other two new submissions ranked in the top 10 for total IOPS. This is the first time we have seen any SPC-1 submissions from TTA, which is a Korean systems company.

There were a number of unusual characteristics of both TTA systems worthy of note.

• They both used 100Gbs InfiniBand attached to the servers running the SPC-1 benchmark.
• They were both AFA using NVMe SSDS but used disks to hold system software (presumably).
• The all HHS2112F flash shelves (5 for the 7-node system and 8 for the 10-node system) were connected to the main controllers via InfiniBand and used NVMe SSDs. It appears as if this connection used NVMeoF/IB.

As far as we can recall there’s been only one other InfiniBand submission in SPC-1 ever and that used direct attached NVMe SSDs for storage (FusionStack FusionStor SF6000). Although the Jet-speed HHS3124F controllers also support flash caching, that was not used for their two SPC-1 benchmark runs.

In Figure 2 we present SCI’s computed, economic performance metric, the top ten IOPS/$/GB

Figure 2 SPC-1 Top 10 SPC-1 IOPS/$/GB

Once again, the 10-node TTA Jet-speed storage did well coming at #5 with 163 IOPS/$/GB. The 10-node system cost of $692K, did 2.4M IOPS and used over 46.8TB of ASU capacity.

As discussed in prior reports, we prefer IOPS/$/GB as a measure of the economics of IO performance rather than the SPC’s $/IOPS (now $/KIOPS). We feel our metric is not biased as much to AFA as the $/IOPS, but this matters less over time, as the only non-AFA system in our top ten in Figure 2 is the (#10) Huawei OceanStor™ 6800 v3.

Next, we show our SCI calculated metric used to measure SSD effectiveness or our top 10 IOPS/SSD ranking. We have used two different metrics to measure SSD effectiveness IOPS/SSD and IOPS/GB-NAND. It remains unclear which provides a better picture of flash IOPS efficiency. We will continue to experiment with both until we determine quantitatively which one is better or we find something that can replace both of them.

Figure 3 Top 10 SPC-1 IOPS/SSD

In Figure 3 we can see both of the TTA submissions performed well as the 7-node Jet-speed came in at #4 32.8K IOPS/SSD (1.5M IOPS/46 SSDs) and the 10-node Jet-speed system came in at #7 with 30.9K IOPS/SSD (2.4M IOPS/78 SSDs). Neither placed in the top 10 for IOPS/GB-NAND.

Finally, we show our top 10 LRT which is the IO latency measured at 10% workload as reported by SPC-1.

Figure 4 Top 10 SPC-1 LRT submissions

Again, both of the TTA Jet-speed systems ranked well here with their 10-node system coming in at #6 with an LRT of 127 sec and their 7-node system coming in at #7 with an LRT of 128 sec, a difference of one microsecond.

Significance

It’s great to see new SPC-1 submissions especially when they break into the top ten charts and new vendors often bring new thinking to storage performance. TTA’s use of InfiniBand for both the host to storage and controller to storage shelf is new to SPC-1 submissions but makes a lot of sense from a storage architecture perspective.

It’s too bad the new Huawei OceanStor 5600 v5 didn’t achieve top ten status in our rankings but it was a midrange system and couldn’t be expected to compete with enterprise class systems.

As always, suggestions on how to improve any of our performance analyses are welcomed.

[This storage announcement dispatch was originally sent out to our newsletter subscribers in February of 2019.  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. Also we have recently updated (May 2019) our SAN Storage Buying Guide so if you are interested in purchasing block primary storage, please checkout our Buying Guide available for sale on our website.]

Silverton Consulting, Inc., is a U.S.-based Storage, Strategy & Systems consulting firm offering products and services to the data storage community

Leave a Reply

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