VMware vSphere

 View Only

  • 1.  RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 04:17 PM

    This picture shows it all..

    I was giving ESX datastore SOME credit, but it appears I was a bit hasty.. the datastore DOES impair performance, here is the proof.

    SAME ESX host, 2 DIFFERENT VDR appliance backups, 2 simultaneous backups. 2 DIFFERENT LUNS on the SAME storage array (Dell MD3200i).

    Both are iSCSI LUNs, but are IDENTICAL in configuration. BOTH LUNS are striped across the same 12 SAS drives (RAID 5)

    The ONLY difference here is ESX01 is RAW, ESX2 is ESX Datastore. The VDR target is a Windows file share on those iSCSI LUNs.

    Look at MAX Kb/s



  • 2.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 07:08 PM

    Was the test done simultaneously or separate times?

    If you found this at all helpful please award points by using the correct or helpful buttons! Thanks!



  • 3.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 07:11 PM

    Was the test done simultaneously or separate times?

    SAME ESX host, 2 DIFFERENT VDR appliance backups, 2 simultaneous backups



  • 4.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 07:15 PM

    I wonder what the stats would show if you did them separately. Makes me wonder if one takes precedence over the other when going simultaneous.

    If you found this at all helpful please award points by using the correct or helpful buttons! Thanks!



  • 5.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 07:21 PM

    I wonder what the stats would show if you did them separately. Makes me wonder if one takes precedence over the other when going simultaneous.

    I tried that, starting one then the other. Makes no difference.. What is interesting about that the stats are the same, so no matter if they run independent or not, the stats for both are the same, that means the datastore (at least for this test) are capped at a certain bandwidth, otherwise the stats should change if it were the only appliance running.

    It also demonstrates the poor disk performance from VDR appliance as a whole.. 44MB/s? That's low. When compared to similar backups, it's quite a difference.



  • 6.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 16, 2010 07:50 PM

    Yeah, that makes sense. Great sample!

    If you found this at all helpful please award points by using the correct or helpful buttons! Thanks!



  • 7.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 18, 2010 04:52 AM

    Hi,

    Nice test, however wondering if you could clarify somethings for me. Looking at the picture it appears that on the LUN ESX01 this is close to 100% reads and the other is almost 100% writes - am I reading this wrong or have I just misinterpreted what you are showing us? It also seems that ESX01 is getting more cache hits that ESX2 so this may explain the differences - why it is getting more would be an interesting question.

    Of course if I'm reading your chart wrong then it makes for very interesting reading indeed.

    Kind regards.



  • 8.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 18, 2010 02:05 PM

    ESX01 is getting more cache hits that ESX2 so this may explain the differences - why it is getting more would be an interesting question.

    I tried to make this post clear and concise rather than include details and too much info because I ASSUMED you would realize that I did all the normal background checking and compared not just based on one test, but this was the LAST test I conducted. Therefore my conclusions were based on a series of tests, not one.

    For clarification sake I will DETAIL what I did, since apparently you don't see the point I am driving at.

    I DID try the switch the LUN's. I did the SAME test the other way by making the VDR point the different LUNs (which is why I chose a Windows file share as the target, so I can make this part easy). I switched the LUNS in the VDR to ensure the VDR themselves were not the choke point or the VM's they were backing up. The results were the same.

    so cache hits would NOT matter since they are IDENTICAL configuration. I didn't just merely run a test and go "HERE look what I found". I didn't post this just to quibble over cache numbers and WRITE vs READ IO. Those numbers reflect how each file system (VMFS vs RAW) handles caching, which could be another indication of why there is a difference in performance.

    the bottom line is DATASTORE doesn't have the same performance as RAW. Doesn't matter how you spin it or change the parameters, it is what it is.. some posts try to defend VMFS (I was one of them) but I see now (which was the entire point of this) I was wrong.

    For people using VDR I thought it might help to improve performance, that was my point. Use RAW as the TARGET in VDR, will dramatically improve performance.



  • 9.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 18, 2010 09:30 PM

    Sorry for asking... and sorry for offending.

    Many apologies. Thanks for your answers.

    Message was edited by: ThompsG



  • 10.  RE: RAW vs ESX Datastore performance - Final Showdown

    Posted Nov 18, 2010 10:09 PM

    Sorry for asking... and sorry for offending.

    Change of heart? Quite different from your original response . . .

    Sorry for asking... ignoring your posts going forward.

    At any rate, consider who is posting and don't assume we don't do research before hand... Given my rating one would think that I didn't just post without cause.