You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Kelly Burkhart <ke...@kkcsm.net> on 2005/04/25 20:21:05 UTC
Re: Windows vs. Linux Performance -- More info
Kelly Burkhart wrote:
> We are attempting to migrate from CVS to SVN and notice very poor
> windows checkout performance compared to Linux. We're running the
> following:
>
> Server:
> Dual Opteron 242 2GB RAM, SuSE 9.1 (32 bit)
> svn 1.1.4 (r13838) (source download, built locally)
> access via svnserve
>
> Linux Client:
> Intel(R) Xeon(TM) CPU 2.66GHz, (HT, 4cpu), SuSE 9.1
> svn 1.1.4 (r13838) (source download, built locally)
>
> Windows Client:
> Intel(R) Xeon(TM) CPU 2.66GHz, (HT, 4cpu), WinXP Pro, SP2
> svn 1.1.4 (r13838) (binary download from tigris.org)
>
>
> I used cvs2svn to convert our cvs repository of about 4500 source files
> in 304 directories to subversion. I did not include branches or tags,
> but I did include trunk history.
>
>
> To checkout trunk w/svn command line app:
> Linux: 0m 15s
> Windows: 2m 53s
I ran iperf between my windows and linux clients (two comparable
machines bouncing off of a cisco switch). Result was 82.6Mbits/sec. I
think the network is fine.
I downloaded Cenatek RAMDiskXP and checked out into the RAM disk. Time
was 1m 35s. I don't know anything about this product other than there
is a free trial download and it must be faster than writing to the
physical disk. (It should be faster than the Linux disk).
How much of an affect would <CR><LF> conversion take? That is the only
work that I can think of that Windows is doing and Linux is not. I
would not expect it to be free, but I can't imagine it dominating the
run time either.
I'm currently creating another repository from the same cvs archive
with: cvs2svn --trunk-only --no-default-eol --keywords-off
If that does what I think it will do, I'll post the results tomorrow.
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Windows vs. Linux Performance -- More info
Posted by Kelly Burkhart <ke...@kkcsm.net>.
steveking wrote:
> Kelly Burkhart wrote:
>
>>> To checkout trunk w/svn command line app:
>>> Linux: 0m 15s
>>> Windows: 2m 53s
>
>
> That's completely expected.
> Subversion builds against the neon library 0.24.7 right now. But that
> version has the HAVE_SETSOCKOPT option not defined on Windows.
> Without that compile option, neon doesn't set some socket options and
> then windows waits a certain amount of time between sent packets.
>
> You can of course compile Subversion yourself, and patch the
> neonconfig.hw file to have a line
> #define HAVE_SETSOCKOPT 1
> in it. That will get rid of that wait time and speed up the whole
> process a lot.
I just took a look at the INSTALL file in the 1.1.4. It seems that neon
only applies if I'm using the web interface. I am using svnserve.
Is there any other reason why building from source may help?
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Windows vs. Linux Performance -- More info
Posted by Kelly Burkhart <ke...@kkcsm.net>.
steveking wrote:
> Kelly Burkhart wrote:
>
>>> To checkout trunk w/svn command line app:
>>> Linux: 0m 15s
>>> Windows: 2m 53s
>
>
> That's completely expected.
> Subversion builds against the neon library 0.24.7 right now. But that
> version has the HAVE_SETSOCKOPT option not defined on Windows.
> Without that compile option, neon doesn't set some socket options and
> then windows waits a certain amount of time between sent packets.
>
> You can of course compile Subversion yourself, and patch the
> neonconfig.hw file to have a line
> #define HAVE_SETSOCKOPT 1
> in it. That will get rid of that wait time and speed up the whole
> process a lot.
I will try building from source.
> I'm not sure if it helps for checkouts if the client is built that way.
> You could try TortoiseSVN, because that one has neon patched with the
> HAVE_SETSOCKOPT option set.
> If it's not any faster, then you have to compile Subversion yourself.
TortoiseSVN (downloaded last week sometime) is slower than the command
line tool.
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Windows vs. Linux Performance -- More info
Posted by steveking <st...@gmx.ch>.
Kelly Burkhart wrote:
>> To checkout trunk w/svn command line app:
>> Linux: 0m 15s
>> Windows: 2m 53s
That's completely expected.
Subversion builds against the neon library 0.24.7 right now. But that
version has the HAVE_SETSOCKOPT option not defined on Windows.
Without that compile option, neon doesn't set some socket options and
then windows waits a certain amount of time between sent packets.
You can of course compile Subversion yourself, and patch the
neonconfig.hw file to have a line
#define HAVE_SETSOCKOPT 1
in it. That will get rid of that wait time and speed up the whole
process a lot.
I'm not sure if it helps for checkouts if the client is built that way.
You could try TortoiseSVN, because that one has neon patched with the
HAVE_SETSOCKOPT option set.
If it's not any faster, then you have to compile Subversion yourself.
Or ask the Subversion devs to either set that option when they do the
Windows build or use neon 0.25 instead (which now has that option set
for Windows itself too).
Stefan
--
___
oo // \\ "De Chelonian Mobile"
(_,\/ \_/ \ TortoiseSVN
\ \_/_\_/> The coolest Interface to (Sub)Version Control
/_/ \_\ http://tortoisesvn.tigris.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Windows vs. Linux Performance -- More info
Posted by Kelly Burkhart <nj...@kkcsm.net>.
Kelly Burkhart wrote:
> How much of an affect would <CR><LF> conversion take? That is the only
> work that I can think of that Windows is doing and Linux is not. I
> would not expect it to be free, but I can't imagine it dominating the
> run time either.
>
> I'm currently creating another repository from the same cvs archive
> with: cvs2svn --trunk-only --no-default-eol --keywords-off
>
> If that does what I think it will do, I'll post the results tomorrow.
Turning off eol translation did not have a significant effect.
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org