Dear All,
Thanks for all your help so far. This page was particularly helpful:
http://www.westnet.com/~gsmith/content/postgresql/chkp-bgw-83.htm
(does the advice for 8.3 apply unchanged to 8.4?)
I'm still hitting issues with this though: sync is taking 7-10 seconds
and I need to get it down to
Dear All,
Thanks for all your help so far. This page was particularly helpful:
http://www.westnet.com/~gsmith/content/postgresql/chkp-bgw-83.htm
(does the advice for 8.3 apply unchanged to 8.4?)
I'm still hitting issues with this though: sync is taking 7-10 seconds
and I need to get it down to
Richard Neill rn...@cam.ac.uk wrote:
So far, I've set checkpoint_segments to 128, timeout to 10min, and
completion_target to 0.8. This helps, but not as much as I'd
hoped.
But I haven't touched any of the other WAL or BG Writer settings.
Where should I look next?
On our web servers,
Richard Neill wrote:
(does the advice for 8.3 apply unchanged to 8.4?)
Yes; no changes in this area for 8.4. The main things performance
related that changed between 8.3 and 8.4 are:
1) VACUUM free space management reimplemented so that the max_fsm_*
parameters aren't needed anymore
2)
On Wednesday 09 December 2009 03:05:40 Greg Smith wrote:
On Linux having the WAL on a separate disk can improve things much more
than you might expect, simply because of how brain-dead the filesystem
fsync implementation is. Reducing the seeks for WAL traffic can help a
lot too.
Not using
Heikki Linnakangas wrote:
I wonder how common this issue is? When we implemented spreading of the
write phase, we had long discussions about spreading out the fsyncs too,
but in the end it wasn't done. Perhaps it is time to revisit that now
that 8.3 has been out for some time and people have
Richard Neill wrote:
On the other hand, the system never seems to be I/O bound. (we have at
least 25 MB/sec of write bandwidth, and use a small fraction of that
normally).
I would bet that if you sampled vmstat or iostat every single second,
you'd discover there's a large burst in write speed
