GET 24/7 LIVE HELP NOW

Announcement

Announcement Module
Collapse
No announcement yet.

XFS for MySQL

Page Title Module
Move Remove Collapse
X
Conversation Detail Module
Collapse
  • Filter
  • Time
  • Show
Clear All
new posts

  • XFS for MySQL

    Hey,
    I'm about to run a new a server, 10x256GB SSD Disks, i plan to use RAID5 (i know it's bad, but that's my only option).
    I will have an adaptec 5445 raid controller with BBU.

    I'm starting to test xfs on some hardware, it's seems as it's faster so i might opt for it.

    i'm planning to create the RAID5 array with 16k chunk size, that makes me mount the xfs with sunit=32,swidth=320,nobarrier
    mkfs.xfs -f -d sunit=32,swidth=320 , am i correct?
    I will use innodb_flush_method = O_DIRECT and innodb_flush_log_at_trx_commit = 1, i should be fine with that?


    Thanks!

  • #2
    raid5 is pretty slow in writes (vs raid10) but still very good in reads. Question is how much write bound is your environment.

    Don't have much experience with XFS but I guess performance would not much differ from ext4. It would be nice to benchmark both with the same data/traffic.

    I'm curious about SSD's - did you test max read/write bandwith ? I wonder if your controller is able to use full SSD drives speed combined.

    Comment


    • #3
      przemek wrote on Mon, 18 October 2010 20:57

      raid5 is pretty slow in writes (vs raid10) but still very good in reads. Question is how much write bound is your environment.


      Do you have a source for that statement when a decent raid controller is used?
      Quote:


      Don't have much experience with XFS but I guess performance would not much differ from ext4. It would be nice to benchmark both with the same data/traffic.

      I haven't seen benchmarks, but XFS is definitely faster.

      Comment


      • #4
        gmouse wrote on Wed, 20 October 2010 14:13

        przemek wrote on Mon, 18 October 2010 20:57

        raid5 is pretty slow in writes (vs raid10) but still very good in reads. Question is how much write bound is your environment.


        Do you have a source for that statement when a decent raid controller is used?


        My statement is a result of my own experiences, but you can find also more info here:
        http://en.wikipedia.org/wiki/RAID_5#RAID_5_performance
        However I agree that decent controller can make a difference, still I doubt you can gain the same write performance as when using raid10 on the same hardware. But feel free to change my mind linking to benchmarks )

        gmouse wrote on Wed, 20 October 2010 14:13


        Quote:


        Don't have much experience with XFS but I guess performance would not much differ from ext4. It would be nice to benchmark both with the same data/traffic.

        I haven't seen benchmarks, but XFS is definitely faster.


        Do you have a source for that statement ? I'd love to read some )

        Comment


        • #5
          przemek wrote on Wed, 20 October 2010 15:29

          gmouse wrote on Wed, 20 October 2010 14:13

          przemek wrote on Mon, 18 October 2010 20:57

          raid5 is pretty slow in writes (vs raid10) but still very good in reads. Question is how much write bound is your environment.


          Do you have a source for that statement when a decent raid controller is used?


          My statement is a result of my own experiences, but you can find also more info here:
          http://en.wikipedia.org/wiki/RAID_5#RAID_5_performance
          However I agree that decent controller can make a difference, still I doubt you can gain the same write performance as when using raid10 on the same hardware. But feel free to change my mind linking to benchmarks )

          Those quotes are old. I don't have the means to perform a benchmark, so I base my opinions on statements as found here (Femme has tried many RAID controllers, not sure if they are db related).

          gmouse wrote on Wed, 20 October 2010 14:13


          Quote:


          Don't have much experience with XFS but I guess performance would not much differ from ext4. It would be nice to benchmark both with the same data/traffic.

          I haven't seen benchmarks, but XFS is definitely faster.


          Do you have a source for that statement ? I'd love to read some )[/quote]
          Mysqlperformanceblog always uses XFS in their benchmarks, and comment #18, even though again the latter is not db related.

          Comment


          • #6
            gmouse: I don't know the language the forum is in, but still I think the discussion makes no much sense until we have some bechmarks comparing database and disk write/read performance on the same hardware using different raid levels.
            From my experience raid5 built on SAS 15krpm as well as on SSD drives had worse write performance than raid10, but I never made any nice charts out of those results. Maybe someday I will have time for this.


            And regarding XFS vs EXT4 there are many benchmarks around, where you can see sometimes better results for one sometimes the other file system, depending on test case and hardware used.
            An example interesting contest I found is here:
            http://www.phoronix.com/scan.php?page=article&item=ext4_ benchmarks&num=2

            My only point is that statement that "XFS is definitely faster" than ext4 is somewhat risky.

            Comment


            • #7
              On the RAID5 vs RAID 10 performance debate I would say that the best comparison that I've read is this:
              RAID10 vs RAID5
              (And the rest in that series).
              But then you have to balance that compared to if put the money that you have to pay for the extra disks compared to if you buy more RAM and/or CPU power instead etc. And that balance is what makes the choice much harder since it depends on the application which is the best way to go.

              przemek wrote on Wed, 20 October 2010 19:21


              My only point is that statement that "XFS is definitely faster" than ext4 is somewhat risky.

              Agree with you fully, all tests that I've read shows that XFS and ext4 are neck to neck when it comes to performance, where some things are a bit faster on one and some on the other.

              Which brings me to my opinion that the underlying hardware decision is a so _much_ bigger issue for performance that the XFS vs ext4 question is really a non-issue.

              Comment

              Working...
              X