Database is on tmpfs and periodically snapshotted to SSD. There are bottlenecks upstream of sqlite that we can see in traces.
On Thu, Aug 22, 2019 at 10:36 AM Warren Young <[email protected]> wrote: > On Aug 22, 2019, at 9:27 AM, Peter da Silva <[email protected]> wrote: > > > > Have an existing application that's pushing the limit > > If the limit is in hardware, shards won’t help. > > For example, a SQLite DB on a 7200 RPM spinning disk is limited to about > 60 transactions per second under the stock SQLite fsync logic, since each > takes 2 revolutions to commit. (One to write to the journal, and one to > commit the journal entry.) Writes to multiple shards only get to share a > platter rotation if there is no seek between writes. > > The limits are higher for SSDs, but there’s still a low limit on the > number of parallel writes. > _______________________________________________ > sqlite-users mailing list > [email protected] > http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users > _______________________________________________ sqlite-users mailing list [email protected] http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
