Thanks Nico, and everyone else. A 3x speed bump is still good, so I've set journal_mode to WAL. The docs seem to suggest I only need to do this once, but I presume that doing it once every time I open the database is acceptable?
Thanks, Ian On 03/06/2011 15:42, Nico Williams wrote: > > On Jun 3, 2011 10:04 AM, "Ian Hardingham" <i...@omroth.com > <mailto:i...@omroth.com>> wrote: > > > > Thank you Igor, I'll do some more thorough profiling. > > > > When I run the query: > > > > UPDATE multiturnTable SET complete=1 WHERE id=-50000 > > > > This takes ~45ms (as reported by SQLite's profile) - is this in the > > right ballpark? I'm running a fairly fast modern intel chip here. > > In journal mode, yes: that's in the ballpark of what three fsync() > calls should take, and that's how many fsync calls SQLite makes at > commit time in journal mode. WAL mode should go faster because most > often it does a single fsync(). But even three times faster will seem > too slow to you, no doubt. To go faster consider batching updates > into larger transactions (so as to amortize the synchronous I/O cost) > or using fast flash devices to hold your DB. Also, serialize your > writing if you can. Finally, consider some other method of obtaining > durability, such as logging updates to a large, widely distributed > memcache farm, then disable fsync for most transactions. > > Nico > -- > _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
