Vincent Fox <vbfox@xxxxxxxxxxx> wrote: > With ZFS there are a number of hits with fdsync bugids. This connects > back to I believe Pascal Giengers' thread, which I think resulted in him > turning off ZIL to bump up performance. This is speculation at this > point, anyhow here's the BugID information for those what's interested: I did not turn off ZIL because I did not want to bump my head out... Turning off ZIL is not a very good idea. You see, every file system can be tuned to be fast turning off all what does guarantee its consistency. Nevertheless I did not want to have a fast mail storage which is completely unreliable ;-) The zil lock problem appeared once but disappeared completely like a miracle. I have reasonable figures now on our servers doing reconfiguration on the storage side, it appeared to be a random i/o problem on storage, and Cyrus is _HEAVY_ random i/o. Not huge data rates but huge seek rates. You can give your storage system some hints, writes do appear in ZFS as sequential ones. Blocks are written in a sequential way (due to COW), but are read in a random way (for sure). ---- Cyrus Home Page: http://cyrusimap.web.cmu.edu/ Cyrus Wiki/FAQ: http://cyrusimap.web.cmu.edu/twiki List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
