subversion performance, one file checkout, no text-base

[Vianney Lecroart <vi...@f4-group.com>]

Hello,

I read mailing list archives and saw that the only messages about 
performance were in 2002. I did some tests and it seems that Subversion 
is faster.
The test server is a GNU/Linux Pentium 3 1Ghz, 512MB RAM on 100MBPS 
Local network (only 12% of the network is used during the test), could 
you confirm that the bottleneck on the server is the CPU frequency? or 
can it be the RAM? I don't really know if I can have better performance 
with a better CPU or with more RAM?

Second question, I saw on the archive that you told about the 
possibility to remove the text-base file on the working copy (at least 
for binary files). Is it implemented? If not, is it planned? When?

Third question, I saw in the FAQ that it's not possible to check out 
only one file. Is it planned to implement it? When?

Thanks,

Vianney

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: subversion performance, one file checkout, no text-base

[Ryan Schmidt <su...@ryandesign.com>]

On Jan 11, 2006, at 09:03, Vianney Lecroart wrote:

> Nobody know these answers?

I think nobody responded because the answers to your second and third  
questions are easy for you to find yourself, and the answer to the  
first question may be somewhat subjective...

> Should I post to the dev list?

If you have specific suggestions for improving Subversion's source  
code, yes. If not, no.


>> I read mailing list archives and saw that the only messages about  
>> performance were in 2002. I did some tests and it seems that  
>> Subversion is faster.
>> The test server is a GNU/Linux Pentium 3 1Ghz, 512MB RAM on  
>> 100MBPS Local network (only 12% of the network is used during the  
>> test), could you confirm that the bottleneck on the server is the  
>> CPU frequency? or can it be the RAM? I don't really know if I can  
>> have better performance with a better CPU or with more RAM?

For what it's worth, the limiting factor on our server appears to be  
disk speed. But our server is not only serving our Subversion  
repository but also ten developers' working copies for a dozen  
projects, along with mail, web and SMB services. The step that  
actually seems to take the most time updating a working copy. Whether  
in our post-commit hook or manually on the command line, this is what  
always seems to thrash the disk and ruin the disk cache.


>> Second question, I saw on the archive that you told about the  
>> possibility to remove the text-base file on the working copy (at  
>> least for binary files). Is it implemented? If not, is it planned?  
>> When?

It is not implemented. The feature request to have working copies  
without the text-base is here:

http://subversion.tigris.org/issues/show_bug.cgi?id=525

This was supposed to be a 2005 Google Summer of Code project; I have  
heard no status update as to whether anything got done.

The enhancement request to have the text-base compressed is here:

http://subversion.tigris.org/issues/show_bug.cgi?id=908

A patch for Subversion 0.31 was supplied in October 2003, but nothing  
appears to have happened since then.


>> Third question, I saw in the FAQ that it's not possible to check  
>> out only one file. Is it planned to implement it? When?

That is correct. It is not currently possible. There is an issue open  
for this:

http://subversion.tigris.org/issues/show_bug.cgi?id=823


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: subversion performance, one file checkout, no text-base

[Vianney Lecroart <vi...@f4-group.com>]

Nobody know these answers?
Should I post to the dev list?

> I read mailing list archives and saw that the only messages about 
> performance were in 2002. I did some tests and it seems that Subversion 
> is faster.
> The test server is a GNU/Linux Pentium 3 1Ghz, 512MB RAM on 100MBPS 
> Local network (only 12% of the network is used during the test), could 
> you confirm that the bottleneck on the server is the CPU frequency? or 
> can it be the RAM? I don't really know if I can have better performance 
> with a better CPU or with more RAM?
> 
> Second question, I saw on the archive that you told about the 
> possibility to remove the text-base file on the working copy (at least 
> for binary files). Is it implemented? If not, is it planned? When?
> 
> Third question, I saw in the FAQ that it's not possible to check out 
> only one file. Is it planned to implement it? When?


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org