When Terrified Vendors Attack: The Dell PowerStore Edition

Dell is at it again. This time, they paid Principled Technologies to do some tests in order to produce a ridiculous report trying to compare the high-end HPE Primera to the midrange Dell EMC PowerStore.

I’ll expose some of the more egregious errors in their methodology and overall thinking, but first I want to direct readers to an easy way to impartially compare for themselves, without having to read a FUD document sponsored by anyone at all.

Executive Summary: A Primera 670 is multiple times faster than a PowerStore 9000T, has stronger data protection, and much higher uptime.

Certified Performance

I can’t come even remotely close to replicating any of their numbers in our (very conservative) sizing tool. Perhaps Hanlon’s Razor is appropriate here. In any case, instead of trying to debug their silly FUD test, I will point you to a public and marketing-free source of performance information.

There is an easy and public way to compare array performance. I thought of this when I saw their ridiculous throughput claims. What better way to discredit Dell than by using their own performance tests that prove PowerStore is many times slower than Primera? 🙂

The SAP HANA storage certification list.

You see, in order to get certified for SAP HANA, a storage system has to pass this incredibly difficult performance test. It is throughput-heavy and requires a system to be able to provide sustained, predictable performance to an SAP HANA cluster. No cheating.

The way to understand the relative performance results between systems is by looking at how many SAP HANA nodes the storage system can reliably feed.

A single HPE Primera 670 with 4 controllers will comfortably satisfy 72 SAP HANA nodes. Even a 2-controller 670 will do 36 SAP HANA nodes.

A single Dell EMC PowerStore 9000T will do 16 SAP HANA nodes.

FYI, Dell in their FUD report was showing a single 7000T as being faster than a single Primera 670 – the 7000T will do 14 SAP HANA nodes, according to Dell.

That is a 4.5x performance delta in favor of the HPE Primera. As in, an HPE Primera 670 is Four And a Half Times Faster than a Dell PowerStore 9000T.

As one would expect from a high end platform, the Primera 670 is a massively more capable system.

So – if the PowerStore is as fast as Dell claims in their FUD document, why isn’t it able to support many, many more SAP HANA nodes than 16?

Ask your Dell FUD-slinger this question, since it’s not something they can hand-wave away with marketing infographics.

[Update Sept 1, 2020: Seems like Dell refreshed their HANA certification page to show numbers with 4 dual-controller appliances. If you run 4x 9000T, then they can do 64 HANA nodes, since each one can do 16 nodes. My analysis is based on a single appliance comparison, just like Dell’s FUD document was based on comparing a single PowerStore system to a single Primera]

Oh –  the Primera achieves this performance with RAID6. The PowerStore uses RAID5. So the Primera is not only multiple times faster, it also has many times the data integrity compared to the PowerStore.

Here is a chart with some more systems for comparison (the data is accurate as of August 24th, 2020). All numbers come from the SAP link above. The numbers are for individual systems. Note that some older Dell EMC systems are faster than the PowerStore 9000T for SAP HANA. Perhaps because they can be set up as RAID10 and without data reduction, whereas the PowerStore can’t use any other RAID than RAID5, and data reduction is forced on.

Here is the PowerStore result. Primera. NetApp. Unity. Pure.

Performance With Reliability

What’s especially sad is that they focused on erroneous speeds and feeds in their report but totally forgot to test things that enterprise customers tend to care about, like data integrity, uptime and performance under adverse conditions.

A far more useful test would be something like this:

  • Pull a controller out. What does this do to performance?
  • Unplug an entire disk shelf. What does it do to availability?
  • Pull out ANY 2 drives simultaneously. Is there ANY chance that the system would stop functioning if you pulled the “wrong” drives? 🙂

You see, a Primera is a high end system, more similar to a Dell EMC PowerMax than it is to a PowerStore. You can do things like configure a Primera with 4 controllers, which are really active-active (as in, all 4 simultaneously will serve a single LUN). You can also do things like set up shelf-level availability, so even the loss of an entire shelf can’t break the system.

Oh – and it offers a 100% uptime guarantee. Which Dell EMC doesn’t offer with any platform at all, not even the PowerMax.

Use Your Head

Why on earth would Dell EMC do something so silly as this FUD exercise?

Perhaps they want to legitimize the PowerStore by comparing it to a high-end array? Or maybe trying to throw shade at the Primera by comparing it to a midrange array? Maybe both?

I will have an absolute field day with the PowerStore RAID implementation, which I’m saving for a separate, special post.

It is a bit weird though that Dell EMC is the last remaining vendor using single parity RAID. Everyone else seems to have graduated to dual parity and beyond. Maybe their performance is broken with dual parity RAID?

Stay safe 🙂

Thx

D

12 Replies to “When Terrified Vendors Attack: The Dell PowerStore Edition”

  1. Great post D. It is indeed sad to see Dell loosing it all together. They should not even talk storage by living in the world of RAID 5.

      1. The node count is a direct result of performance achieved. If Dell EMC believes the PowerStore 9000T can really do more than 16 nodes, they’re more than welcome to certify extra nodes. Somehow I don’t think Dell couldn’t find enough servers

        The PowerMax with the max 16 controllers does a very healthy node count, for example. So clearly their HANA certification labs have a plethora of nodes to play with.

        At this point, the only public and marketing-free measure of performance for these systems is the SAP HANA node count.

        Vendors want to certify as many nodes as possible since that directly affects the SAP HANA economics.

        1. You said it… Not enough servers available. So not a measure of performance. I still don’t agree counting SAP nodes is a measure of performance. Where’s the math in that?

          1. Or not more than 16 nodes required to run workload. If your unit of measure is nodes, you’ll have to assume all node configurations the same and transactions achieved the same per node. As you know, with in-memory databases node config. matters. Dell could argue they only needed 16 nodes to achieve workload measure.

          2. LOL no, I said the opposite, that Dell has tons of servers to play with, as proven by their PowerMax numbers. I will prove my point with a reductio ad absurdum example.

            If, like you say, the unit of performance is NOT servers, then a low end system that is certified for only 1 server can run the same amount of workload as a system that is certified for 1000 servers? Because maybe that 1 server is insanely fast and the 1000 servers are from 30 years ago?

            Are you SURE that makes sense to you? 🙂

            No, the entire point of this is that it provides a standardized measure of performance for consumers of HANA, and the test itself is audited.

            It’s the entire point of the certification, to provide a standard measure of comparison for HANA!

            Every vendor strives to get the absolute highest node count certified since that makes them more competitive.

            And 16 nodes isn’t a bad number, it’s just not 72 🙂

            To put it in perspective, the Nimble AF80 (similar class system as the PowerStore 9000T) is also certified for 16 nodes.

            However, the AF80 does 16 nodes with:

            – one controller active (the other is standby)
            – with Triple+ RAID, not RAID5.

            Pull a controller out of the 9000T and imagine it running triple parity RAID and it would be lucky if it managed a handful of nodes.

            Which also means the per-controller node count is a nice measure of controller performance density and efficiency, and the 9000T is lacking (especially given the RAID5 scheme).

            So, to summarize, the certified SAP HANA node count IS a standardized measure of performance, and everyone wants to get the absolute max number.

            It’s funny how Dell updated their certification after my post, to show the score for FOUR 9000T running together. Four boxes can do 64 nodes. But they still have to show the score for the single system, and that remains 16 nodes.

            Of course, 4 systems would cost 4x what one 9000T does, and still not achieve the 72 nodes a SINGLE Primera 670 does.

            It was one of the design goals of the Primera to squeeze a ton of performance density per U, and the goal was eminently achieved as you can see.

            Hopefully after this long explanation it’s finally clear.

            And none of this is helped by the fact that Dell is still using RAID5 in the PowerStore. It’s 2020, no other major vendor is using single temporally correlated URE protection!

            Thx

            D

  2. To the point, backed up by facts, and from reliable sources! Something we can all appreciate! A new name, and a new faceplate does NOT count as innovation…oh, and single parity RAID5? That is actually a step BACKWARDS, not forward!

  3. Great insight and response D.. perfect befitting answer to a baseless FUD of Dell with just some Random Numbers published on a Paid Website .. Looking forward to more Ammo for a Frontal attack Plan….No defense anymore..Thanks.

Leave a Reply

Your email address will not be published. Required fields are marked *

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