On Wednesday 12 July 2006 14:14, Magnus Hagander wrote: > > there is a way to 'cheat' : either use fsync = off in > > postgresql (that sucks), or try a writeback enabled raid controller. > > In both cases you're taking a risk, but it's rather low in > > the second one. > > That's not entirely correct. > fsync=off is pretty much the same thing you have with MyISAM, so if > you're fine with that, you shuld be fine with fsync=off. > > However, if you use a writeback enabled RAID controller (*with* a > battery backup of course - if you have one without a battery, someone > ripped you off), you do *not* take a bigger risk. You will get almost > the performance of fsync=off without any real risk. This is really the > only option for *any* database system if you want both performance and > stability. I've allready experienced this "without risk" writeback cache firsthand, when the controller crashed... yes, that happens too... and then you can be sure your data is corrupted (very corrupted...). Of course, it's really rare. It hapened only once for me, but there was a lot of damage.
> > > These two cheats both remove the sync wait created by the > > transactions on the 'WAL' (journal) files on postgresql. > > No. fsync=off removes *all* fsync, but the WAL does not necessarily use > it. WAL can also use OPEN_SYNC, OPEN_DSYNC etc, in which case it's > really not affected by fsync=off. However, all data file access during > checkpoints and such things also fsync, so you'll definitly get more > performance (at a risk) by turning it off. But this part is good to know :) Anyway thanks a lot for correcting my mistakes :) ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
