Diamond Notes

Just another WordPress weblog

New Benchmarks Coming Soon — Your Chance to Participate!

February 13th, 2008 | Category: Benchmarking

I am going to be getting a new master/slave pair of production servers. I will have between a one and two weeks before they have to go into production.  This will be a great chance to do a little benchmarking. I am going to begin writing up what I plan on testing in the next few days. I would love to here your suggestions about what to test!!

Here is the specs as I know them:

8 gigs of RAM

4 or 6 disks — 10k or 15k drives on a RAID controller with 256MB write-back cache.

4 or 8 cores (cpu) — most likely eight though

debian etch

MySQL 5.0.45

Things I want to test:

write-back turned on vs write-back turned off on RAID controller

xfs/reiserfs/ext3 on the RAID partition

I/O schedular (debian defaults to CFQ — want to test deadline also)

innodb-flush-log-at-trx-commit = 0 vs 1

I will be using our QPP (query processing programs) tools to run tests with production data and production client requests.

I very likely will not have time to run all the tests I want. However, the more tests the merrier. Please comment with your test ideas.

7 Comments so far

  1. Freddy February 13th, 2008 7:01 pm

    I would be interested in a LVM-Benchmark. We are planing to use LVM on our servers for better backup capability (snapshots), but we ar enot sure about the performance-impact.

  2. admin February 13th, 2008 7:24 pm

    Thanks for the input. We use lvm on all our servers for the very reason you mention. My understanding that the impact is very minimal - on the order of a couple of percent at most. Since we use it on all machines already and that won’t change it won’t be something we test. Ideally I would like to, but we just aren’t going to have enough time.

  3. Kevin Burton February 13th, 2008 8:10 pm

    Testing with deadline is sort of a requirement.

    Also… Try to test with MySQL 5.1 and a large innodb_thread_scalability value (2x # cores).

    Also, try to run sysbench with SELECTs with high numbers of threads and your entire DB in memory.

    I bet you become CPU bound.

  4. Nils February 14th, 2008 12:59 am

    I’d like to see benchmarks of different RAID Levels and Stripe sizes

  5. David Holoboff February 14th, 2008 9:00 am

    I am interested to find out what your multi-thread results are for xfs/reiserfs/ext3 with the two schedulers you mentioned (CFQ and deadline). We’re doing a couple of test in that area in the next few days…

  6. Brice February 17th, 2008 3:21 pm

    Hi,

    I’m also doing some benchmarks (sysbench oltp large db and sql-bench) of mysql over ext3 or XFS on a 4 disks RAID10 array with 512MB BBU ram cache.

    I’m really disappointed by XFS performance in comparison to ext3 (with XFS being about 25% slower than ext3). I cared to tune XFS per the recommendation I could find on the xfs list.

    I’m wondering if you’ll find the same results as I do. Please post the results for comparison.

    Another thing to test is to disable the read cache if you can on your raid card. Let the kernel or mysql cache the read data, and keep most of your raid cache free for writes.
    Also, if you have time, try different stripe size (and beware of partition alignment: http://insights.oetiker.ch/linux/raidoptimization.html).

  7. Diamond Notes » Benchmark Updates March 9th, 2008 6:23 pm

    […] I blogged about recently I was going to have a chance to do some testing.  Last week we started benchmarking our two new […]

Leave a reply