Admin Spotting for Fun and Profit

3ware performance in CentOS

June 13th, 2008 Posted in Linux

There is an upstream bug report here, which may be of interest to folks using 3ware and aacraid raid cards with CentOS or RHEL. The basics of the bug hinge on setting MWI, which can have a pretty hefty impact on IO performance. If you’re storing your MySQL bits on a 3ware powered array, it’s a safe bet that this fix may help improve your performance and reduce some of the IO wait seen on the system.

The downside with this is that even though the fix is known, Red Hat is sticking to their procedure, and has stated that they will not release the fix for this in the main kernel until 5.3 is released. Since 5.2 is fresh from the factory, it’s not likely that we’ll be seeing this fix pushed mainstream in the next few months.

This leaves a few choices for the RHEL and CentOS communities for how to proceed.

  1. Weigh in with your opinion on this bug report. If enough people respond, RH will likely appease them.
  2. Help test the patched kernels in the bug report. The more comfortable RH is with the patch, the more likely it is that they’ll tuck it in with a bug fix or security update. See #1.
  3. Give CentOS a chance to get 5.2 out the door. Once it’s released, folks will have some time to roll up a kernel repository at http://people.centos.org similar to what was done with bz321111. Since this bug is strictly performance affecting, installs won’t be an issue, and you can update to the modified kernel, or use the stock release as you see fit.

In theory, you could also roll your own kernel with this patch if you didn’t want to wait, however if you do this, you’re accepting responsibility for building it properly and tracking all the kernel security and bug updates until the patch becomes mainstream. I wouldn’t recommend this method since it requires more time and upkeep, but for folks who roll their own it provides another alternative.

  1. 9 Responses to “3ware performance in CentOS”

  2. By Larry Ludwig on Jun 13, 2008

    Can you post the URL of this bug for more info?

  3. By Jim Perrin on Jun 13, 2008

    Should be in the body of the article, but sometimes the colors I use make it hard to see links. Here it is:
    https://bugzilla.redhat.com/show_bug.cgi?id=444759

  4. By pirast on Jun 14, 2008

    god. i dont unterstand the intention of the red hat policy in such a case…

  5. By Laurent Wandrebeck on Jun 16, 2008

    Nice to see this bug is gaining a bit of importance.
    Thanks for making a post about it Jim.

  6. By Matt on Jun 16, 2008

    Sorry if this is a silly question, but would this only affect 3ware and aacraid raid performance on CentOS 5.x or would this same issue affect the 4.x kernel?

    Thanks.

  7. By Jim Perrin on Jun 17, 2008

    I don’t really think that’s silly at all. I hadn’t really thought about it since I’ve migrated everything I run to centos5 now, but it’s entirely likely that it applies. The libata bits have changed around a little but it’s an interesting notion. Something for you to contribute to the upstream bug maybe?

  8. By Laurent Wandrebeck on Jun 18, 2008

    Matt, as far as I can remember, that “bug” has been fixed in 2.6.22 dev cycle. So, if your bios is buggy, you’ll very likely hit that problem with 4.x.

  9. By Sean Keeney on Jul 9, 2008

    Hi,

    Thanks for pointing out this problem, but does someone have a link please to a fixed kernel? I looked at the centos site and saw lots of custom kernels in the link above but no indication of which one I need. Is there a repo I could use?

    Thanks,

    Sean

  10. By Jim Perrin on Jul 9, 2008

    Johnny Hughes was nice enough to create an area for them on people.centos.org

    The direct link is here -> http://people.centos.org/hughesjr/kernel/5/bz444759/

    These may not improve your performance however. I ran a timed ‘dd’ style test with and without the kernel, and the differences I experienced were negligible. Others have reported drastic improvements, so your mileage may vary.

Post a Comment