Re: Transactionless reporter performance results

I missed this back when you posted it, but better late than never right?

Greg Hudson wrote:
>
> I'm not sure why there is a ~14-second variance in some of the trials.
> I'd say it has to do with BDB logfile creation, but the new code
> shouldn't result in any logfiles being created.
>

There are generally 2 things that cause massively fluctuations in my
performance results.

1) Background Processes - When I am done writing my test scripts and
have gotten some results, I switch to single user mode with almost
nothing else running in the background and do the same tests there.
Generally the results will be extremely close but the single user mode
results will have less deviation in them.

2) Disk Caching - Before I run the command about to be timed, I remount
the partition and run the sync command once to empty the disk cache.

Your modified script might look like this:

rm -rf wc ; cp -r wcsave wc ; umount wc; sync; mount wc; cd wc ; time
svn up ; cd ..

The only difficulty with running this test is if you didn't leave a
small amount of disk space unpartitioned to be able to umount the
filesystem in question.

p.s. I have not found a way to clear out the disk cache on Windows
without rebooting. If anyone knows how to achieve this, I'd love to
hear about it.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Jan 30 04:55:10 2004