You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Christopher Baus <ch...@yahoo.com> on 2003/10/04 08:11:13 UTC

Windows co very slow timeout then fail.

I have a RedHat 9.0 server running 0.30.  I spent
about a week setting it up and importing my code.  I
just finished importing a large project, and from a
Windows 2000 client did:

svn co http://ip/rep test

The windows client becomes unresponsive and eventually
times out.  It literally can take minutes to get one
small file.  The linux client is rather slow, but it
never takes more than about a second to get the file. 
I've seen some issues about performance posted to the
list, but I haven't seen any regarding only the
windows client.  Is this a known issue?  

I have to admit this is rather frustrating after
taking so much time to learn the system and create my
repositories.  I'm honestly out about a week's work if
I can't get the windows client working.

Sincerely

Christopher Baus.



__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: Windows co very slow timeout then fail.

Posted by "C. Michael Pilato" <cm...@collab.net>.
Ben Collins-Sussman <su...@collab.net> writes:

> "C. Michael Pilato" <cm...@collab.net> writes:
> 
> > Ben Collins-Sussman <su...@collab.net> writes:
> > 
> > > Can anyone else reproduce this?  I mean, if I run the win32
> > > commandline client against a large Linux-apache server, is it likely
> > > to happen to me?
> > 
> > Ben, you'll recall that I've been doing simultaneous checkouts of
> > /trunk on a Windows box from a Linux/Apache server.  Ten checkouts
> > going at once, all pounding my little NTFS harddrive, and they all
> > completed successfully in about 9 minutes.  At one file per second, we
> > would be able to checkout only 540 files.  But we do soooo much better
> > than that -- there are over 2000 files in our /trunk tree, spread
> > across some 150 directories.  That means that with *ten* checkouts
> > hitting the same server and pounding the same client's disk, we're
> > doing more like 3-4 files per second.
> 
> Ahhhh, right.   Thanks.
> 
> So the question is: what is it about Christopher's hardware/software
> or environment that is causing this behavior, since it seems *not* to
> be the norm?

FAT32?  Network-aware anti-virus software (checking every file as it
comes down, every tempfile as it gets made, etc.)?  I dunno, dude.


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

Re: Windows co very slow timeout then fail.

Posted by Ben Collins-Sussman <su...@collab.net>.
"C. Michael Pilato" <cm...@collab.net> writes:

> Ben Collins-Sussman <su...@collab.net> writes:
> 
> > Can anyone else reproduce this?  I mean, if I run the win32
> > commandline client against a large Linux-apache server, is it likely
> > to happen to me?
> 
> Ben, you'll recall that I've been doing simultaneous checkouts of
> /trunk on a Windows box from a Linux/Apache server.  Ten checkouts
> going at once, all pounding my little NTFS harddrive, and they all
> completed successfully in about 9 minutes.  At one file per second, we
> would be able to checkout only 540 files.  But we do soooo much better
> than that -- there are over 2000 files in our /trunk tree, spread
> across some 150 directories.  That means that with *ten* checkouts
> hitting the same server and pounding the same client's disk, we're
> doing more like 3-4 files per second.

Ahhhh, right.   Thanks.

So the question is: what is it about Christopher's hardware/software
or environment that is causing this behavior, since it seems *not* to
be the norm?


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

Re: Windows co very slow timeout then fail.

Posted by Jan Evert van Grootheest <j....@euronext.nl>.
Christopher,

Perhaps you just run into the same issue I had....
Apparently, the subversion *server* scales pretty bad with the number of 
files in a directory. I had one directory with over 5000 files in it 
(ACE documentation) and there were problems during checkin and checkout.
A local checkin succeeded, but took more than 15 minutes after the 
'transmitting files' stage. A remote checkin succeeded even though the 
client reported a connection timeout (the server kept busy about the 
same 15 minutes).
Using the repository after this large checkin was terrible.
Checking out that directory would succeed locally (file:///), but fail 
remotely (http:/). The local checkout would start at some 3 or 4 files 
per second and slow down to 1 file every 5 or so seconds. Really unusable.

The first thing to check is whether the client or the server is busy 
chewing. If its the server, perhaps you have the same issue. If it is 
the client, it might well be a virus scanner or other things that have 
already been suggested.
I use tortoiseSVN on windows and sometimes it jsut takes a long time to 
show the proper directory contents. That really is due to the virus 
scanner, because the directory contains MS word documents. And also 
windows' taskmanager shows that the virus scanner is really very busy.

-- Jan Evert

Christopher Baus wrote:

> So I'll take it there is something wrong with my
> setup.  I get the same results now on both WindowsXP
> and 2000.  I had authz running, but read that it could
> slow down performance.  I disabled it.  No luck.  Also
> disabled HTTP authentication.  No luck.  
> 
> I have the wxWindows tree in my repository.  That
> gives you an idea of the size.  I'm currently running
> from the binaries, but I would build a debug client if
> necessary.  I suspect that is just going to be blocked
> on the socket read code.  If anyone else has any
> suggestions, please let me know.  I need to get this
> working.  Now it is eating away my weekend.
> 
> Thanks,
> 
> christopher
> 
> __________________________________
> Do you Yahoo!?
> The New Yahoo! Shopping - with improved product search
> http://shopping.yahoo.com
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
> 


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

Re: Windows co very slow timeout then fail, more info...

Posted by Christopher Baus <ch...@yahoo.com>.
Here's another piece of information.  I didn't do
anything on my XP laptop and now it is working with
0.30.0 on that plaform..

Here's summary of what happened.

Install 0.30.0 on RedHat 9.0 with latest apache2.

Install 0.30.0 from distribution on Windows XP/2000
clients.

Create new repository

Set HTTP authentication and authz.

import basic directory structure.

co on windows 2000/XP and linux.  Works fine.

Import wxWindows and makefiles into repositry.
co on Windows 2000 -- fails with timeout getting wx
directory.
co on Windows XP -- fails with same error

remove authentication
co on Linux -- works
co on Windows still fails

back client to 0.29.0 on Windows 2000.
co goes slowly and hiccups, but then works.
co again with 0.29.0 works fine.
co with 0.30.0 works fine.

co on WindowsXP with 0.30.0 works fine.

Strange.  I don't have the foggiest what happened, but
it nows looks like the problem involved both the
client and the server, as I didn't change my client at
all on XP and it started working after using 0.29.0
client on 2000.  

I just don't have to investigate this further, but it
does bother me.


__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: Windows co very slow timeout then fail, more info...

Posted by Christopher Baus <ch...@yahoo.com>.
Ok this is really weird...

For kicks I backed my client to 0.29 and it worked.  I
was convinced something is getting hosed up in the
0.30.0 Windows distribution, but then I tried 0.30.0
and it worked also.

I was able to complete the checkout.  Is anyone using
the 0.30.0 Win32 client distribution against large
repositories?  I'm starting to feel that I'm the only
one using 0.30 on reasonable sized repositories, and
that scares me.  I want to confirm this bug.  It is
really nasty because it seems intermittent.  I wonder
if had to something to do with authz?  But that might
be a red herring.  

I still don't think performance is all that great.  If
SVN is going to replace CVS, the performance has to
get MUCH better IMHO.  

Anyway.  Its working again.  I'll cross my fingers and
hope I don't see this behavior again, as I'm pretty
much committed to SVN now.

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: Windows co very slow timeout then fail, more info...

Posted by Ben Collins-Sussman <su...@collab.net>.
Christopher Baus <ch...@yahoo.com> writes:

> I'm pretty convinced that either the client or server
> are frivilious blocking.  CPU, disk, and network usage
> on both is virtually nil.  

What happens when you create a 'local' repository on the windows box,
import some stuff, and try to use the same win32 cmdline client to do
a checkout of file:///  ?

Is it just as slow?

If so, it makes me wonder whether libsvn_wc is acting up in your
client.

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

Re: Windows co very slow timeout then fail, more info...

Posted by Christopher Baus <ch...@yahoo.com>.
I'm pretty convinced that either the client or server
are frivilious blocking.  CPU, disk, and network usage
on both is virtually nil.  


__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: Windows co very slow timeout then fail.

Posted by Christopher Baus <ch...@yahoo.com>.
So I'll take it there is something wrong with my
setup.  I get the same results now on both WindowsXP
and 2000.  I had authz running, but read that it could
slow down performance.  I disabled it.  No luck.  Also
disabled HTTP authentication.  No luck.  

I have the wxWindows tree in my repository.  That
gives you an idea of the size.  I'm currently running
from the binaries, but I would build a debug client if
necessary.  I suspect that is just going to be blocked
on the socket read code.  If anyone else has any
suggestions, please let me know.  I need to get this
working.  Now it is eating away my weekend.

Thanks,

christopher

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com

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

Re: Windows co very slow timeout then fail.

Posted by "C. Michael Pilato" <cm...@collab.net>.
Ben Collins-Sussman <su...@collab.net> writes:

> Can anyone else reproduce this?  I mean, if I run the win32
> commandline client against a large Linux-apache server, is it likely
> to happen to me?

Ben, you'll recall that I've been doing simultaneous checkouts of
/trunk on a Windows box from a Linux/Apache server.  Ten checkouts
going at once, all pounding my little NTFS harddrive, and they all
completed successfully in about 9 minutes.  At one file per second, we
would be able to checkout only 540 files.  But we do soooo much better
than that -- there are over 2000 files in our /trunk tree, spread
across some 150 directories.  That means that with *ten* checkouts
hitting the same server and pounding the same client's disk, we're
doing more like 3-4 files per second.

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

Re: Windows co very slow timeout then fail.

Posted by Ben Collins-Sussman <su...@collab.net>.
Christopher Baus <ch...@yahoo.com> writes:

> I have a RedHat 9.0 server running 0.30.  I spent
> about a week setting it up and importing my code.  I
> just finished importing a large project, and from a
> Windows 2000 client did:
> 
> svn co http://ip/rep test
> 
> The windows client becomes unresponsive and eventually
> times out.  It literally can take minutes to get one
> small file.  The linux client is rather slow, but it
> never takes more than about a second to get the file. 
> I've seen some issues about performance posted to the
> list, but I haven't seen any regarding only the
> windows client.  Is this a known issue?  

I've never heard of this.  You're saying that the Linux client is
receives a file about every second;  that sounds like normal, known
HTTP performance for svn 0.30.

But I have no idea why the windows client would behave any
differently... it's the exact same code.  I assume you're talking
about the plain old svn 0.30 win32 commandline binary.

Can anyone else reproduce this?  I mean, if I run the win32
commandline client against a large Linux-apache server, is it likely
to happen to me?

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