Percona's xtrabackup utility has a --throttle option to reduce the IO load of the backup job.

The docs say that the value passed is the number of read/write pairs per second. Is 1000/sec an appropriate value on modern hardware? How about 5? I cannot find any meaningful frame of reference for these values.

For reference - I am reading from and writing to the same drive array (10k SAS). A 55gb backup job with --throttle=20 ran in roughly an hour with no apparent strain on the system during off-peak hours. But I honestly don't know if this is a high or low value for the throttle.

1 Answer 1


it all depends but you can guesstimate something with this approach: take a look at iostat and look at iops/sec on your disk. if you have typical database you're most probably limited by number of random seeks/sec, not the bandwidth.

  1. in maintenance window - run xtrabackup without throttling and see again what numbers of iops/sec can your system generate. say it's x.
  2. after that look how many iops/sec is typical for system during off-peak hours. say it's y.

based on this make some estimation how many iops/sec can you dedicate to backup job. i would calculate it as x - 2 * y or x - 3 * y to leave some headroom for spikes.

i think xtrabackup's parameter will be linearly proportional to iops/sec but not equal - so in last step use trial and error to tune throttle value so iostat shows you desired number of operations/sec.

alternativly use ionice [ a little about it here ], give your backup job low priority and do not throttle it at all. i'm doing it for rdiff-backup jobs - works quite nicely. note that ionice [ afaik ] works only with some io schedulers in linux.

  • Good idea, thanks. I've tried ionice previously for this sort of thing but have never been able to get my desired results.
    – allaryin
    Jun 6, 2009 at 6:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.