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

Re: status unbearably slow on large repositories

From: Shurik O <shuriko_at_gmail.com>
Date: 2004-11-19 02:17:49 CET

Ben,

>> The question is: what are you expecting, realistically? Certainly CVS
>> would take just about as long, at least in the same ballpark. Maybe
>> you shouldn't run 'svn status' at the very top of your working copy, but
>> at selected subdirs further down.

It might be same ballpark for you but I found SVN at least 2 times slower in my quick comparison -- see below.
The repositories were 80% the same (big, ~20K files). Actually, it *feels* more than 2x slower. Please also note
that the CVS command I used, cvs -n update does more than svn status -- it goes to the server to check if there are
newer updates there.

For us converting to from CVS to (excellent in any other respect) SVN may turn out to be quite a challenge.

Regards,
Shurik

--------- CVS ----------------------------------
timeit cvs -nq update
...
Version Number: Windows NT 5.1 (Build 2600)
Exit Time: 4:50 pm, Thursday, November 18 2004
Elapsed Time: 0:01:43.578
Process Time: 0:00:04.281
System Calls: 1240617
Context Switches: 570346
Page Faults: 42081
Bytes Read: 14875749
Bytes Written: 42683453
Bytes Other: 6804181

-------- SVN ------------------------------------
timeit svn status
...

Version Number: Windows NT 5.1 (Build 2600)
Exit Time: 4:55 pm, Thursday, November 18 2004
Elapsed Time: 0:03:34.796
Process Time: 0:00:16.703
System Calls: 3228827
Context Switches: 1409927
Page Faults: 193374
Bytes Read: 284276602
Bytes Written: 12366523
Bytes Other: 14658257

Ben Collins-Sussman wrote on 11/18/2004 4:38 PM:
>
> On Nov 18, 2004, at 3:28 PM, Arni J Rognvaldsson wrote:
>
>> I'm trying to convert a largish vss repository to svn. As I'm sure you
>> understand, I can't stand working with vss. Not everybody feels this
>> way though, the lack of locking is a major roadblock, but the time svn
>> status takes on the repository is a show stopper. I'm using version
>> 1.1.1 on Windows, among other platforms. The repository has 44k files
>> in it, excluding svn's own files (202k of those...). Running svn
>> status takes about 5 minutes, which is just way too long. I understand
>> there may be a fix for this that did not make it into the 1.1.1
>> version? If so, is there a 1.1.2 version in the works anytime soon? I
>> would build it from source but the Windows build instructions look
>> daunting.
>>
>
> There was a speedup patch committed on November 1st, but I don't think
> there's any patch that will make a difference to you. You're asking
> windows to stat() 44,000 files to look at timestamps, at a bare
> minimum. That's going to take -minutes- long, no matter how much you
> optimize away other disk i/o.
>
> The question is: what are you expecting, realistically? Certainly CVS
> would take just about as long, at least in the same ballpark. Maybe
> you shouldn't run 'svn status' at the very top of your working copy, but
> at selected subdirs further down.
>
> It may also just be a hindrance to have a working copy that big; 'svn
> update' and 'svn commit' are going to have to crawl over the whole
> thing, doing at least as much disk I/O as 'svn status'. Maybe you need
> smaller working copies of separate modules.

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Fri Nov 19 03:02:03 2004

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