kfogel wrote:
> Hal Mahaffey <hmahaffey@aol.com> writes:
>> already had a case where a single directory contained build
>> manifests for 15,000 builds. Performance accessing these files (in
>> CVS) was noticeably painful. We solved it by indexing the files in
>> subdirs based on the 1st character of the build name: A...Za...z.
>>
>> Has anyone noticed performance problems with repositories with
>> 10,000+ revisions?
>
> I don't believe we've seen any actual reports of it being a problem.
>
>> What "older native filesystems" have this problem? We were using
>> Veritas v3.5 on EMC Symms and noticed the problem. Are there any
>> recommended filesystems?
>>
>> If there *is* a problem with larger repositories, has the Subversion
>> community considered indexing the 'db/revs' directory with, say, the
>> first digit (so revision 2423 would go in db/revs/2)? Just a
>> thought.
>
> Yep, we've considered that and other similar approaches. We decided
> not to do anything until we hear reports of the current arrangement
> being a problem, that's all.
What a wise person!
I wouldn't expect problems with ext3 filesystems, even with 2.4 Linux
kernels, until you hit better than 5000 files in a directory. Certain
operations, such as "ls -l" operations bog down quite badly: 2.6 kernels
fixed this by implementing "htrees" for ext3.
Your Mileage May Vary for other filesystems: Reiserfs is famous for being
able to handle things like news spools, but I find it unstable in extended
use.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Fri Mar 31 04:12:36 2006