back to article NetApp embiggens E-Series flashbox: Gee, a benchmark... thanks

NetApp has doubled the speed and capacity of its all-flash E-Series arrays, and won an SPC-1 benchmark record in the process. The new EF560 is an E-Series all-flash array that builds on the existing 25-month-old EF550 to raise its sustained IOPS number from the 400,000 or so at EF550 launch time in November 2013 to past 650, …

  1. Bucky O' Hare

    "Embiggens"

    There's that word again. It's a creeping menace.

    1. Anonymous Coward
      Anonymous Coward

      Re: "Embiggens"

      Embiggens is a perfectly cromulent word

      - the simpsons

  2. Anonymous Coward
    Anonymous Coward

    Damning with faint praise

    Story illustrated with a picture of the world's slowest "High Speed Train"...

  3. Anonymous Coward
    Anonymous Coward

    $/Performance/Latency?

    I'd have like to see a charts based on $/Performance and latency, vs just rank on IOPs. Do these things exist somewhere? They would be more meaningful, imo.

    Cheers

  4. Anonymous Coward
    Anonymous Coward

    Is that a NetApp spin win? Second place is a win?

    Summary:

    "NetApp has doubled the speed and capacity of its all-flash E-Series arrays, and won an SPC-1 benchmark record in the process."

    But in the article:

    "the best rating of any array in SPC-1 test history except for a Kaminario all-DRAM array's $0.40/IOPS"

    I guess second place isn't good enough for NetApp? Come on Chris.

  5. dikrek
    Stop

    The low latency is the star here

    Hi everyone, Dimitris from NetApp here.

    The EF560 result was meant to be impressive in 2 ways:

    1. Low latency for the money

    2. High IOPS at a low latency

    Maybe this article will clear it up:

    http://recoverymonkey.org/2015/01/27/netapp-posts-top-ten-spc-1-price-performance-results-for-the-new-ef560-all-flash-array/

    Thx

    D

    1. Archaon

      Re: The low latency is the star here

      The graphs are admittedly somewhat small to read but as far as I can see you've achieved performance that just about noses past a 5 year old T-class 3PAR, and pricing that looks be be on par with a modern 3PAR 7400 array?

      Well done, I guess?

      1. dikrek

        Re: The low latency is the star here

        look at the latency. The key is - what are the SPC-1 IOPS at each latency point?

        1. Archaon

          Re: The low latency is the star here

          "look at the latency. The key is - what are the SPC-1 IOPS at each latency point?"

          Uhm, still better than the NetApp - more IOPS and lower latency. And that's bearing in mind that it's the older version of the 7400 (admittedly the upgrades aren't huge on the newer one). And even then it's not the SSD-optimised model (7450).

          Anyway I'm by no means saying it's rubbish because Array X is 2% quicker than yours but it's probably best to play to your strengths. The results are good for a NetApp system but it's hardly a whitewash against the other vendors. I may be misinformed but it's my understanding that NetApp boxes never play particularly well in all-out block performance because of the emulation of block storage on top of file storage?

          Play to your strengths, don't try and take on other people's strengths and pretend you do them better. Still with the good old lines "It does <insert NetApp marketing here>" but use it as a feather in your cap that you can end the paragraph with "And look, HP will tell you our block performance is rubbish because we emulate it on top of our file layer but look how our SPC-1 performance is within 3% of theirs while offering you all of <insert NetApp marketing here again> that HP cannot do with their arrays."

          1. dikrek
            FAIL

            Re: The low latency is the star here

            Sorry Archaon - the E-Series is as pure a block array as they come. You may want to familiarize yourself with our offerings.

            Do visit my article - at 500 microseconds, the EF560 does 196K SPC-1 IOPS at $0.68/op and the 3Par does 129K IOPS at $1.15/op. http://bit.ly/18oWI1R - look at the table towards the end.

            My point is - it depends what you want. Functionality? then All-Flash FAS offers more than anyone else at great speeds. Crazy low latency? Then EF has everyone beat for the money.

            Thx

            D

            1. Anonymous Coward
              Anonymous Coward

              Re: The low latency is the star here

              "at 500 microseconds, the EF560 does 196K SPC-1 IOPS at $0.68/op and the 3Par does 129K IOPS at $1.15/op"

              Sorry but once again you're cherry picking the data against an older result from a hybrid array running a less cost effective SSD variant. You still haven't managed to surpass that system in real terms on this latest generation AFA, so the obvious conclusion is that you're attempting to put a much more positive spin on this Netapp result than was really achieved based on actual $/SPC-1 IOPs rating,

          2. Anonymous Coward
            Anonymous Coward

            Re: The low latency is the star here

            This particular Netapp/LSI box has none of those features you mention, there is no block emulation on file, this is a 100% block array so no file system creating additional latency. The performance result is pretty good, not so the price given it's being compared to SSD pricing in a hybrid array from way back in mid 2013. The other problem is that this really is a pretty basic array with limited features vs both Netapp legacy FAS and it's AFA & Hybrid competitors.

            Looking at the SPC report, do submissions no longer need to account for sparing anymore ? It seems to be missing from this and a few of the more recent results I've seen. Maybe I'm misreading the report but surely you wouldn't deploy any system without spares and that overhead would most definitely impact the overall purchase cost and ultimately the $/IOp score awarded by SPC.

            1. DLow

              Re: The low latency is the star here

              Hi Daniel from Netapp here (with the usual disclaimer of this being me and not the company speaking)

              In regards to Spares in SPC tests (and in general) - Its up for debate/depends imho.

              Many of the SPC1 tests are with RAID10 and a smaller number of disks involved (per group/total) hence the need for a spare is not critical. And as SPC1 is a write heavy test using RAID10 is of course a performance play too, no raised eyebrows there I think.

              Here we are also talking about SSD's where the failure rate is much lower from the start compared to HDD. And on top of that its small disk sizes too, 400GB in our test but SSD's today (in the enterprise offerings) are no bigger than 1.6TB.

              So smaller amount, smaller capacity and much faster transfer rates make for shorter rebuild times and time exposed.

              Now, if you want spares, close to zero exposure and can live a bit less performance I would suggest looking at our DDP technology. 800GB SSD's can be back into “safe mode” in as little as 15 minutes.

              And I know I said DDP gives less performance but in the flash world this is all relative, ~175,000 IOPS at 400us (75/25 8k OLTP) when using DDP isnt half bad.

              Anyway, to spare or not to spare? Maybe it comes down to (customer) preference and how you trust your systems. Personally have not suggested or designed E/EF based solutions with dedicated spares for some time now as I, and more importantly, our collected sensor data says we don’t really have to.

              1. Anonymous Coward
                Anonymous Coward

                Re: The low latency is the star here

                "So smaller amount, smaller capacity and much faster transfer rates make for shorter rebuild times and time exposed."

                Granted but you still need somewhere to rebuild to otherwise you're exposed until the disk is physically replaced so not the kind of system you would really recommend to a Customer even with a dual parity scheme in place. In this case it was potentially even worse because Raid 10 was used for the benchmark.

                So either you're reserving space within the pool in case of a failure, which means that portion of the pool isn't really usable and that's not showing in the report or you need to add dedicated spares and neither is that.

                Either way that should count as unusable space on a benchmark or Customer system and that will have an affect on the overall economics and the subsequent SPC report since it will either push the price up or lower the usable capacity.

                I suspect spare space was intentionally left out of the configuration for this very reason, or are you trying to suggest spares (whether reserved space or disk) just aren't needed for flash systems ?

            2. bitpushr

              Re: The low latency is the star here

              There is no such thing as "block emulation on file". There is also no such thing as "file systems [creating] additional latency". Good grief.

              Disclaimer: I am a NetApp employee.

              1. Anonymous Coward
                Anonymous Coward

                Re: The low latency is the star here

                "There is no such thing as "block emulation on file". There is also no such thing as "file systems [creating] additional latency". Good grief."

                Really ? is that why a Netapp LUN is actually a file sitting in the WAFL file system which then sits on block, if that's not emulation, what is ? Added layers e.g LUN --> Volume --->,WAFL ---> Block adds latency. If it didn't you'd just sell the FAS without all the features instead of having a completely different and incompatible architecture for block.

  6. Man Mountain

    We're getting beyond the point where customers are buying flash for pure performance these days so benchmarks and performance POCs are largely pointless. Flash is fast, we get it! But functionality whilst being bloody fast is the battleground now. It's very rarely a pure drag race these days, and for the small number of customers where absolute raw ridiculous performance is key, they already know where to look. There's not many workloads a few hundred 1000 IOPS at sub ms latency won't swallow up, but how easy is it to live with the array?

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like