On 12/05/2012 09:22 AM, Keith Chew wrote:
On Wed, Dec 5, 2012 at 1:14 PM, Keith Chew<keith.c...@gmail.com> wrote:
Since there is no fsync (I know this using grep on the strace ouput),
it is hard to tell what to look for in the strace. The output is very
noisy, so it makes it hard to go through them.
Does anyone know any handy flags for strace to only see all the
activities associated with sqlite file access?
I have confirmed using strace there is no fsyncs done by sqlite (using
synchronous=0 or synchronous=1), just seeks and writes. fsync is seen
only when synchronous=2.
After conducting more tests (using synchronous=0), I have some results
which is baffling me:
- Mysql average insert time, with write-cache OFF: 1-2ms
- Mysql average insert time, with write-cache ON: 0-1ms
- sqlite average insert time, with write-cache OFF: 160ms (have seen
up to 4000ms)
- sqlite average insert time, with write-cache ON: 50ms (have seen up to 150ms)
Quite a significant difference. I am stumped on how to make sqlite run faster.
If it's not fsync() then IO delays are normally caused by read().
You could try [strace -T -eread ...] to check.
Are SELECT statements fast on the same database? How large is the
database compared to the machines memory?
Dan.
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
