Intel SSD 910 in tpcc-mysql benchmark

I continue my benchmarks of Intel SSD 910, the raw IO results are available in my previous experiment. Now I want to test this card under MySQL workload to see if the card is suitable to use with MySQL.

  • Benchmark date: Sep-2012
  • Benchmark goal: Test Intel SSD 910 under tpcc-mysql workload and compare with baseline Fusion-io ioDrive card
  • Hardware specification
    • Server: Dell PowerEdge R710
    • CPU: 2x Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz
    • Memory: 192GB
    • Storage: Fusion-io ioDrive 640GB, Intel SSD 910 (software RAID over 2x200GB devices)
    • Filesystem: ext4
  • Software
    • OS: Ubuntu 12.04.1
    • MySQL Version: Percona Server 5.5.27-28.1
  • Benchmark specification
    • Benchmark name: tpcc-mysql
    • Scale factor: 2500W (~250GB of data)
    • Benchmark length: 2h, but the result is taken only for last 1h to remove warm-up phase
  • Parameters to vary: we vary innodb_buffer_pool_size: 13, 25, 50, 75GB to have different memory/data ration. And we test it on two storages: Fusion-io ioDrive and Intel SSD 910
  • Results
    There is graph of Throughput taken every 10 sec:

    Jitter graph:

    Or to have final results I take total amount of transactions for 1h:

    BP sizeFusion-ioIntel SSD 910Ratio (fio/i910)
    13 GB3971573527501.13
    25 GB7240114977691.45
    50 GB146655911242231.30
    75 GB246413519394151.27


    In conclusion I see that Intel SSD 910 handles MySQL workload quite well, I did not face any problem working with this card.
    Level of stability of results is about the same as with Fusion-io card. The performance of Intel SSD 910 is about ~30% worse, but
    it is expected for this price level. I think Intel SSD 910 is suitable to use with MySQL / Percona Server.

    Link to raw results and stats
    Raw results, config, OS and MySQL metrics are available from Benchmarks Launchpad.

    Share this post

    Comments (9)

    • Peter Zaitsev


      Very interesting…. it for some reason with smaller BP size when workload is expected to be more IO bound the difference from FusionIO is smaller than for less IO bound workload which is very counter intuitive. Do you have any ideas why it would happen?

      Also did you do similar run for conventional Hdds… I wonder how much faster is it…

      September 7, 2012 at 10:37 am
    • Andy

      Is 910 crash safe – does it have capacitors to save data?

      September 7, 2012 at 10:47 am
    • Vadim Tkachenko


      I am not totally sure, however document
      says that card can return SMART attribute “Volatile Memory Backup failure” which corresponds to broken capacitor.
      So I guess it has.

      September 7, 2012 at 11:28 am
    • Vadim Tkachenko


      Yes, it is not obvious, but I have a theory.

      As you see form raw IO benchmark, Intel 910 is much better in reads than writes.

      Having smaller BP corresponds to more reads intensive workload, so that’s why Intel shows less gap there.

      I am going to have the similar benchmark against RAID10 to compare numbers.

      September 7, 2012 at 11:31 am
    • Daniël van Eeden

      Hi vadim,

      What about different firmware levels?

      September 10, 2012 at 10:15 pm
    • Vadim Tkachenko


      I am not sure I understand your question.

      September 11, 2012 at 7:54 am
    • Daniël van Eeden

      Hi Vadim,

      Different firmware releases can have a big impact on the performance of the fusion io devices. Did you try different firmware’s? And could we expect anything from a new firmware?

      September 11, 2012 at 12:41 pm
    • Vadim Tkachenko


      I tested it with Fusion-io Drivers 3.1.5. Fusion-io does not really support downgrading version, so I can’t test with
      older releases. I know that 3.x series for Fusion-io provides better performance than 2.x one.

      September 11, 2012 at 1:56 pm
    • Joakim Sernbrant

      | MySQL Version: Percona Server 5.5.27-28.1

      The logs suggests that this was on MySQL-5.6.6, not 5.5.

      February 6, 2013 at 9:52 am

    Comments are closed.

    Use Percona's Technical Forum to ask any follow-up questions on this blog topic.