[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

RE: Speeding up workspace

From: Bert Huijben <rhuijben_at_sharpsvn.net>
Date: Sat, 7 Feb 2009 01:42:23 +0100

> -----Original Message-----
> From: Toby Thain [mailto:toby_at_telegraphics.com.au]
> Sent: Friday, February 06, 2009 11:52 PM
> To: Ruslan Sivak
> Cc: Bert Huijben; users_at_subversion.tigris.org
> Subject: Re: Speeding up workspace
> On 6-Feb-09, at 4:28 PM, Ruslan Sivak wrote:
> > Toby Thain wrote:
> >>
> >> Switch to Linux or OS X, which should be good for an immediate
> >> 10-20x speedup.
> >>
> >> --Toby
> >
> > Do you have any benchmarks to back this up?
> >
> Benchmarks were recently posted here by Jonathan Taylor (showing 10x
> speedup in client operations on Linux/ext3 versus Windows), I'd be

Just to prove that these numbers say nothing without a thorough further

The average 1 file operation (say revert) will be more than ten times faster
on NTFS than on ext3 in Subversion 1.6.

Reason: For working copy stability Subversion has to wait until the
timestamp on the filesystem changes, to make sure we always detect file
changes (same timestamp+same size = file unmodified).
The ext3fs has a timestamp granularity of 1 second, while NTFS (and e.g.
EXT4) have submicrosecond precision.

Subversion 1.6 will try to detect these more advanced filesystems in the
time otherwise spend on waiting and when it detects and advanced filesystem
it continues without waiting.

It's not a big improvement for the CLI (where it only waits when finished
the combined operations), but users of subversion GUIs that perform multiple
operations in one batch will certainly notice the difference.

So: please don't use Subversion (with its 2 layers of platform
compatibility) as a benchmark for the filesystems itself. (Of course you can
use it as benchmark for Subversion on that filesystem, but that is not the
same thing)

Subversion is a very VERY bad example for testing actual filesystem
performance with all those compatibility layers.

And like I said in my previous mail: Please show us(/me) where the real pain
is with actual testcases (preferably real world use cases). I think there is
a lot of improvement possible with the current working copy implementation
and especially with the new working copy for 1.7.

> happy to make new benchmarks (OS X &/or Linux/reiser3) if somebody is
> really interested and can't make their own. I have a non-trivial real
> world repository to test with.

I'm checking out at our @WORK repository at a /not recommended/ level above
many projects and all of their branches to create a really big working copy
testcase.. Let's see what the profiler says on checkout/update/status of
this working copy.

My AnkhSVN (and @WORK) colleague Sander will probably create a Linux /
Windows comparison on similar Virtual Machines over the weekend for some
other tests. So let's see what we can find.



To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_subversion.tigris.org].
Received on 2009-02-07 01:45:05 CET

This is an archived mail posted to the Subversion Users mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.