Steve Lord wrote:
On Fri, 2003-03-07 at 08:30, Andy Skunza wrote:
Stephen Lord wrote:
Possibly the way your app runs is making things worse. There is a script
in the cmd/misc directory called xfsstats.pl you can watch log activity
with this. In particular xs_log_writes and xs_log_blocks. If log_blocks
is not a reasonable multiple of log_writes then you are doing a lot of
very small writes.
What is a "reasonable multiple" of log_writes to log_blocks? Would 22 be
unreasonable? And, how to use xfs_stats.pl per xfs partion?
Thanks,
andy
The stats are global and not available on a per fs basis. You can clear
the stats with
echo 1 > /proc/sys/fs/xfs/stats_clear
So you can measure more controlled periods of time.
Perfection is 64, the more recent your code the closer you should get,
we have been removing synchronous transactions from the code. My
workstation which has been up 27 days is averaging a ratio of 52.
It also depends on what you run, something which does O_SYNC writes
all the time can hammer it.
So, for a workstation that has been up for 6 days (running relatively
new XFS from CVS) seeing a value of around 20 to 21:1 is not there
yet.
Do you want us to monitor this for you and provide feedback?
(This is a development system - that is, build/compile/edit/etc,
and not a server - I will check on those later)
--
Michael Sinz -- Director, Systems Engineering -- Worldgate Communications
A master's secrets are only as good as
the master's ability to explain them to others.
|