On 12/3/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > "P Kishor" <[EMAIL PROTECTED]> wrote: > > I get 1000+ inserts a second for a random 100 byte string > > insert ( > > I get 50000+ inserts/sec on my Linux box.
turned off AutoCommit, turned off rand string generation, and lookee what I get... 200,000 inserts in half a second. If my math is not wrong, that works out to more than 360 thousand inserts per sec on the same laptop with a 5400 rpm 2.5" disk. Of course, I batched up the commit at the end. Including the COMMIT, the speed dropped down to almost 67,000 inserts per sec. sqlite> select count(*) from foo; 200001 sqlite> All of the below comes from a wiser man than I. > > Insert speed is not the issue. It is COMMIT speed. At > each commit, SQLite waits until all data is on oxide before > continuing. That will typically take at least two rotations > of the disk platter, or about 17 millisecond, depending on > your disk drive. Waiting for data to get to oxide is > part of being ACID. You can set: > > PRAGMA synchronous=OFF; > > and your COMMITs will go *much* faster because it will no > longer wait on the disk drive. But if you lose > power in the middle of a commit, you might corrupt your > database file. > > Note that there is an implied BEGIN...COMMIT around every > INSERT statement if you do not explicitly start a transaction > using your own BEGIN. > > -- > D. Richard Hipp <[EMAIL PROTECTED]> > ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
