On Monday 11 October 2010 19:23:46 Mladen Gogala wrote: > On 10/11/2010 10:02 PM, Scott Carey wrote: > > Did you tune the linux FS read-ahead first? You can get large gains by > > doing that if you are on ext3. blockdev --setra 2048<device> > > Actually, I have blockdev --setra 32768 > > > would give you a 1MB read-ahead. Also, consider XFS and its built-in > > defragmentation. I have found that a longer lived postgres DB will get > > extreme file fragmentation over time and sequential scans end up mostly > > random. On-line file defrag helps tremendously. > > I agree, but I am afraid that after the demise of SGI, XFS isn't being > developed. The company adopted the policy of using only the plain > vanilla Ext3, which is unfortunate, but I can't do much about it. There > is a lesson to be learned from the story of ReiserFS. I am aware of the > fact that Ext3 is rather basic, block oriented file system which doesn't > perform well when compared to HPFS, VxFS or JFS2 and has no notion of > extents, but I believe that I am stuck with it, until the advent of > Ext4. BTW, there is no defragmenter for Ext4, not even on Ubuntu, which > is rather bleeding edge distribution. When it comes to a database that has many modifications to its tables, it seems that XFS pulls way ahead of other filesystems (likely) because of its on-line defragmentation among other reasons. I am not sure that XFS is not (properly) maintained. The people at xfs.org seem to be making steady progress, and high quality updates. I have been using it for some time now (9+ years), and as it is, it does everything I need it to do, and it is very reliable. I really can't see anything changing in the next few years that would effect its usability as a filesystem for Postgres, so until something proves to be better, I can't understand why you wouldn't want to use it, maintained or not. -Neil- -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
