On Wed, 2003-03-19 at 01:12, Neil Harris wrote:
I've been trying to profile the system to find the
bottleneck.
It's those damn users! Server was running *just fine* until people tried
to use it. ;)
Some simple disk monitoring showed very high levels of
disk activity,
both read and write, yesterday. This is surprising, as the site should
be able to run mostly out of memory, except for writes to update the
database when pages are edited.
Mysql's key buffer is only 384M. (Much more than that with our current
setup and we'd be in much bigger danger of thrashing the swap.)
* Can someone who knows the code tell me whether there
is much use of
intermediate files in the Wikipedia software?
Session management involves temporary files in some way, I don't know if
they're written frequently. Everything has log files of course which are
written out constantly -- Apache and the wiki software each write a line
of log for every page fetched. Mysql's binary log makes a note of every
update, insert, and delete. Every page view increments the page view
counter in the stats table (a single row, very small, very fast). The
Innodb transaction log file is too small and is probably flushing more
often than it needs to (I had troubles when I tried to set it larger --
it told me it was the wrong size, naturally, and refuse to ran!)
That's several writes per page fetch; at circa 4 pages per second I'd
expect on the order of 20-50 writes per second. 250 does sound a little
high...
* Can someone who knows the hardware tell me how many
drives are in the
RAID, and what RAID level is used?
As far as Linux can see, there is a single drive, an IBM DDYS-T36950N
(36704 MB). The SCSI channel is in 160 MB/sec mode (card is an Adaptec
AIC-7899). If there's any RAID, I don't know about it.
-- brion vibber (brion @
pobox.com)