On Thu, Dec 6, 2012 at 3:52 PM, Keith Chew <keith.c...@gmail.com> wrote: > I am guess it may have something to do with the locking of the files. > sqlite could be getting blocked by the filesystem when trying to > obtain the lock before writing to the file...
This is a stab in the dark... I had a look at the sqlite code, and noticed that the file lock is acquired using fcntl(F_SETLK). So, when kjournald wants to flush the data to disk, and sqlite wants to write to the file at the same time, I guess there will be contention, and sqlite can only get the lock when the file is free. This explains the latencies that I am seeing being consistent with the ext3 journal's parameters. Eg increase commit time removes the contention (thus 0ms between commits), removing write-cache, disabling barriers all affect the latency accordingly. So, if this is the case, how do I get around it? Regards Keith _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
