NetApp system zips past IBM monolith in all-flash array benchmark scrap

NetApp has beaten IBM's biggest, baddest all-flash storage box in an industry-standard benchmark. IBM's DS8888 scored 1,500,187 IOPS on the SPC-1 v3 benchmark, while NetApp's A800 did 2,401,171 IOPS – 60 per cent faster. The SPC-1 benchmark tests shared external storage performance in a test designed to simulate real-world …

[Stephen McLaughlin]

These Numbers are Just Insane

Seems it wasn't all that long ago getting a sustained 100K out of a storage unit was considered acceptable performance.

[dan_isaacs]

Re: These Numbers are Just Insane

The IOPS are nice, but the latency....

2.4M SPC-1 IOPS @ .590ms

~2M SPC-1 IOPS @ ~.400ms

Crushes any other result, so far. And that's with regular Gen5 FC fabrics. Imagine what Gen6 32GB with NVMe/FC will do to bring latency down even further. :)

[MikeShea]

Re: These Numbers are Just Insane

And it still is, for most workloads - unless of course you don't want to ever wait for your data - and no one does! Spinning disk would do the job on my laptop, but I'd never do that again.

[2Nick3]

Cheating the numbers?

I wonder if anyone has "VW-ed" their array microcode to recognize it is being tested for a SPC-1 benchmark and do something to cheat the results.

[Justicesays]

Re: Cheating the numbers?

"I wonder if anyone hasn't "VW-ed" their array"

FTFY

The SPC benchmarks are massively gamed and always have been.

Commercially nonsensical hardware optimized specifically to the benchmark.

Loads of controllers with tiny amounts of disk each to maximise the cache memory available.

Volumes made from just the fastest part of traditional disks , while still quoting the full capacity in the $/GB.

Specialized firmware setups are just the tip of the iceberg.

[dan_isaacs]

Re: Cheating the numbers?

Hi Justice,

The details of the tested config are available in the FDR

http://spcresults.org/sites/default/files/files/full_disclosure_report/A32007_FDR.pdf

You can see there that each pair of nodes used the new AFF A800 w/24 NVMe drives and 16Gb FC. DataONTAP 9.4 was the array OS, which any customer can download. While you aren't wrong that many vendors will try to game the test, NetApp has an extensive history of testing only configurations our customers would use, or in this case, are already using.

Be mindful to prevent skepticism from becoming cynicism. :)

Mahalo,

Dan Isaacs

[Anonymous Coward]

Re: Cheating the numbers?

Ah yes, the Dieselgate of storage. Performance that you'll never actually get in real life unless your workload happens to have the same working set demands as SPC-1.

Of course it does help that vendors run their own tests.

[J. Cook]

That's why I prefer the traditional 'bench mark' test...

... which is to drop it on the bench from a distance of 1 meter above the top of the bench, and observe the results- what type/size of divot is impressed on the bench, and if the equipment is still usable afterwards...

[Secta_Protecta]

Re: That's why I prefer the traditional 'bench mark' test...

I did see a rack containing a NetApp 3050 and shelves fall out the back out of a truck once back in the day. 1 metre + drop onto concrete and it still powered up afterwards :)