Обсуждение: freebsd/softupdates for data dir
Hi, reading the archives i cant find a clear answer about softupdates in freebsd, is it recommended to enable it for the data directory? --- miguel
Miguel wrote: > Hi, reading the archives i cant find a clear answer about softupdates in > freebsd, is it recommended to enable it for the data directory? > There is a pretty good article about softupdates and journelling here: http://www.usenix.org/publications/library/proceedings/usenix2000/general/full_papers/seltzer/seltzer_html/index.html and in the freebsd docs here: http://www.freebsd.org/doc/handbook/configtuning-disk.html Postgres does not do a lot of file meta-data operations (unless you do a *lot* of CREATE/DROP INDEX/TABLE/DATABASE), so the performance gains associated with softupdates will probably be minimal. I've always left them on, and never had any issues...(even after unscheduled power loss - which happened here yesterday). As I understand it, the softupdate code reorders *metadata* operations, and does not alter data operations - so the effect of fysnc(2) on a preexisting file is not changed by softupdates being on or off. Cheers Mark
On Apr 3, 2006, at 10:10 PM, Mark Kirkwood wrote: > I've always left them on, and never had any issues...(even after > unscheduled power loss - which happened here yesterday). As I > understand it, the softupdate code reorders *metadata* operations, > and does not alter data operations - so the effect of fysnc(2) on a > preexisting file is not changed by softupdates being on or off. This is also my understanding, and I also leave softupdates on for the data partition. Even if it doesn't improve performance, it will not reduce it, and otherwise does no harm with respect to postgres' disk usage.
On Apr 4, 2006, at 10:41 AM, Vivek Khera wrote: > On Apr 3, 2006, at 10:10 PM, Mark Kirkwood wrote: > >> I've always left them on, and never had any issues...(even after >> unscheduled power loss - which happened here yesterday). As I >> understand it, the softupdate code reorders *metadata* operations, >> and does not alter data operations - so the effect of fysnc(2) on >> a preexisting file is not changed by softupdates being on or off. > > This is also my understanding, and I also leave softupdates on for > the data partition. Even if it doesn't improve performance, it > will not reduce it, and otherwise does no harm with respect to > postgres' disk usage. More importantly, it allows the system to come up and do fsck in the background. If you've got a large database that's a pretty big benefit. -- Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com Pervasive Software http://pervasive.com work: 512-231-6117 vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461
On Apr 5, 2006, at 6:07 PM, Jim Nasby wrote: > > More importantly, it allows the system to come up and do fsck in > the background. If you've got a large database that's a pretty big > benefit. That's a UFS2 feature, not a soft-updates feature.
On Thu, Apr 06, 2006 at 09:45:34AM -0400, Vivek Khera wrote: > > On Apr 5, 2006, at 6:07 PM, Jim Nasby wrote: > > > > >More importantly, it allows the system to come up and do fsck in > >the background. If you've got a large database that's a pretty big > >benefit. > > That's a UFS2 feature, not a soft-updates feature. It's both. You can't background fsck with softupdates disabled. -- Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com Pervasive Software http://pervasive.com work: 512-231-6117 vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461