This Storage Intelligence (StorInt™) dispatch covers Storage Performance Council (SPC) results . There have been no ESRP submissions in years now so in in its stead, we offer our latest SPC-2 performance report. Since our last report on SPC-2, over a year and half ago, there have been 2 new submissions, the Telecommunications Technology Association (TTA) SCALEWAY SG1000-UF and NetApp EF600 AFA.
As you may recall the primary performance metric for SPC-2 is MBPS, the average max bandwidth over SPC-2 workloads. As such, we begin our discussion with top 10 SPC-2 MBPS™ performance in Figure 1.
Figure 1 Top 10 SPC-2 MPPS
The new NetApp EF600 AFA came in at our new #9. It’s performance here is significant because of it’s the first, 32Gbps NVMe/FC system result we have for SPC-2. There was one other system that used NVMe SSDS (Vexata VX100-F #4) but in a standard FC attach. All other systems in our top ten used SAS connected SSDs or disk drives. For some reason, the Vexata submission is no longer active but I retain it here as it had very good performance.
NetApp’s EF600 AFA is an end-to-end NVMe solution, which supports 32Gbps FC NVMe/FC all the way up to 200Gbps NVMe/IB or NVMe/E-RoCE. The EF600 is a 2U dual controller, rack mounted system which had 64GB of cache, 20 1.9TB NVMe SSDs, and 16-32Gbps NVMe/FC host connections.
In Figure 2, we show the top 10 systems in SPC-2 $/MBPS or Price-Performance.
Figure 2 SCI derived Top 10 SCI SPC21 $/MBPS Price Performance
The new NetApp EF600 solution came in as our new #1 at $3.53/MBPS just ahead of its older sibling, the EF570 AFA at $3.69/MBPS. Coming in at #4, is the new TTA SCALEWAY SG1000-UF AFA at $6.67/MBPS.
The TTA SCALEWAY SG1000-UF also had 20 3.8TB SSDs but they were SAS connected. The SCALEWAY solution had 32GB of cache and used 4 16Gpbs FC host links. Note, the SCALEWAY only supported one controller. That may have boosted its price performance. Unclear whether the SCALEWAY SG1000-UF offers a dual controllers solution or not, but it’s unlikely to be deployed in the enterprise without some sort of an HA solution.
In Figure 3, we present our top 10 LDQ-LFP-VoD spider chart which breaks out the performance of the three bandwidth intensive workloads in the SPC-2 MPBS benchmark.
The SPC-2 doc’s I have (SPC v1.7 2017) say that the
• LFP (large file processing) has 3 workloads run in sequence, 100% write, 50:50 read-write and 100% read at 1024KiB and 256 KiB data transfer sizes. Which is probably why LFP is the hardest one to do well in. read.
• LDQ (large database query) has one workload simulating large scans or joins of database at 1024KiB and 64KiB data transfer sizes. As far as I can tell this is mostly (99%) read activity.
• VoD (video on demand) has one workload and represents a video streaming workload also 100% read activity. The doc I have doesn’t indicate the data transfer size but our guess is 256KiB
So one key to doing well on SPC-2 benchmarks is to perform large reads very well. See how the top systems perform on each separate workload can shed some insight into what they have focused their systems performance on.
Figure 3 SPC-2 Top 10 LDQ-LFP-VoD spider chart
As we showed in Figure 1, the new NetApp EF600 AFA came in as our new #9 in MPBS but Figure 3 shows how well it did on the 3 distinct SPC-2 workloads, LDQ, LFP and VoD, that are averaged to generate the SPC-2 MBPS.
Most of the other systems on the chart seem to do really well in one or at most, two of the workloads (LDQ and VoD, typically) but the (inactive) Vexata VX10[0-F and Kaminario K2 did surprisingly well on all 3. The NetApp EF600 seemed to perform all three workloads well enough but LDQ (39.3GB/s) was best, VoD (30.7GB/s) was next and LFP (23.3GB/s) was worst. In contrast the #1 Fujitsu ETERNUS DX8900 S3 system did 841.GB/s, 73.7GB/s and 52.6GB/s, respectively
It’s great to see new SPC-2 submissions especially when they rank well on some of our top ten charts. Yes, the new systems didn’t do that great in MPBS but managed to do very well in price performance. Even with little write activity in SPC-2, the fact that an AFA NVMe/FC solution can do well in the SPC-2 bandwidth benchmark speaks volumes as to how well the industry has managed to a) improve SSD overall read throughput and b) mask SSD asymmetric (reads well, writes poorly) performance.
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 July 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