On Thu, 19 May 2011 21:45:23 -0500 "Leslie Rhorer" <lrhorer@xxxxxxxxxxx> wrote: > OK, I may have made an error, and I want to make sure before this > goes any further. Yesterday I created a RAID6 volume with eight 3TB > members, on the same machine containing an older RAID6 array built of > fourteen 1G members. Based upon my reading here and some level of > experience with larger stripe sizes, I decided to create the array with a > chunk size of 4096K. 4096K is an absurdly large chunk for such RAID5/RAID6 array, I wanted to refer some benchmarks, but couldn't even find any that would test such a huge chunk, largest they test is 1024 or 2048, where the write performance already drops off very sharply, e.g. see: http://blog.jamponi.net/2008/07/raid56-and-10-benchmarks-on-26255_10.html http://louwrentius.blogspot.com/2010/05/raid-level-and-chunk-size-benchmarks.html While the read performance does seem to not suffer or even increase at higher chunk size, the best read/write combination performance seems to be achieved at 64...256K chunk for RAID5/6. If you decide to persist with the current configuration, make sure you increase stripe_cache_size by a lot (and btw raising it if you have the RAM to spare is a good idea in any case). -- With respect, Roman
Attachment:
signature.asc
Description: PGP signature
