This Storage Intelligence (StorInt™) dispatch covers Storage Performance Council (SPC) results. Since our last report, two quarters ago there have been 4 new SPC-1 v3 submissions, three from Fujitsu, ETERNUS DX900 S5, AF150 S3 and the AF650 S3 and one from Telecommunications Technology Association (TTA) Gluesys AnyStor-700EK. All these submissions were AFA. There were also 4 new SPC-2 submissions of the same systems. We will cover SPC-2 results in a future performance report. This report will focus on these systems SPC-1 results.
We begin our discussion with top 10 SPC-1 LRT™ performance in Figure 1.
Figure 1 Top 10 SPC-1 LRT
The new TTA Gluesys AnyStor-700EK came in as our new #8 with ~135 µsecsand the new Fujitsu ETERNUS DX900 S5 came in an #10. The Gluesys storage used 24 480GB SAS/SATA SSDs and 100Gbps InfiniBand networking. Recall, that LRT is measured at 10% of peak workload, or 31K IOPs not a heavy load these days so that may have helped them to do so well. Nonetheless, ~135 µsecs is pretty damn fast for in my view.
The Fujitsu ETERNUS DX900 S5 used 192 960GB SAS/SATA SSDs and was measured at 160K IOPs. This is a more substantial workload and subsequently seems more realistic. Also 151 µsecs is also pretty damn fast, especially for standard 16Gbps FC interfaces.
In Figure 2 we present SPC-1 Price-Performance™ metric, the top ten $/K-IOPS.
Figure 2 SPC-1 Top 10 SPC-1 Price-Performance results
The new Gluesys AnyStor-700EK came in as our new #2 with ~$84.0/ K-IOP. They had a relatively small configuration 4TB ASU capacity and only had 1 InfiniBand connection, which made their system inexpensive for the 310.5K IOPs they were able to achieve.
The Fujitsu ETERNUS AF150 S3 came in at our new #3 with $89.40/K-IOP. The AF150 was also a smaller system which only had 5.5TB of RAID 1 ASU capacity and 2 FC ports which led to another inexpensive system for the 280K IOPS they were able to achieve.
Next we display a SCI derived bubble chart which shows the types of IOPS and LRT attainable for AFA and disk only systems in the SPC-1 results.
Figure 3 SCI derived chart IOPS X LRT with bubble size=$/GB for systems
As we can see in Figure 3 there are definitely many AFA systems with LRT below 1msec and most disk only systems range in LRT from 1.3 to 3msec in SPC-1 results. It’s interesting to note that there are a gaggle of AFA, relatively inexpensive (in $/GB) systems capable of 800K IOPs or more and there is a wide cost ($/GB) range of AFA systems that are capable of < 500µsec LRTs.
It’s great to see new SPC-1 and SPC-2 submissions especially when they rank in some of our top ten charts. It’s interesting to see a relatively inexpensive system achieve top 10 LRT results but it’s unclear the advantage that came with InfiniBand host interface. There aren’t many InfiniBand results in SPC-1 data (only 3 TTA systems in our top 30 LRT results use InfiniBand).
I know saving SPC-2 results is unkind but given the paucity of new SPC submissions or for that matter any performance benchmark results in recent quarters I felt it best to spread them out across two reports.
As always, suggestions on how to improve any of our performance analyses are welcomed.
This storage performance was originally sent out to our newsletter subscribers in February of 2020. 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. ]
Silverton Consulting, Inc., is a U.S.-based Storage, Strategy & Systems consulting firm offering products and services to the data storage community
 Summary metrics for the TTA Gluesys AnyStor-700EK were not available at press time but we were able to interpolate an approximate LRT of ~135µsecs, from a graph in the Full Disclosure Report. We may be off by +/- 5µsecs which could change its rankings but it would still be somewhere between #7 to #9 in LRT.
 Unclear when SPC-1 stopped using $/IOPS and went to $/K-IOPS but it makes sense and I will start using it from now on to be consistent with SPC