You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Ari S <sa...@gmail.com> on 2009/09/29 13:53:43 UTC
Comparison Collabnet and VisualSVN
Dear all,
Currently, we are investigating a migration from CVS to Subversion.
For the Subversion server, we have compared specs of different
suppliers, and the shortlist has now been narrowed down to two:
CollabNet and VisualSVN Server.
When performing performance tests with the two, we found CollabNet to
be about 30% faster than VisualSVN server when using "out of the box
configurations" (as little changed as possible).
Has anyone else found similar (or totally different) figures?
Are there other pros and cons that should be considered in choosing
between these two vendors?
Could anyone perhaps point me to such performance (or feature)
comparison investigations done in the past by others?
Any information would be greatly appreciated.
Ari S.
Leiden, The Netherlands.
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2401635
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Mark Phippard <ma...@gmail.com>.
On Wed, Sep 30, 2009 at 7:31 AM, Ivan Zhakov <iv...@visualsvn.com> wrote:
> On Wed, Sep 30, 2009 at 3:25 PM, Mark Phippard <ma...@gmail.com> wrote:
>> On Wed, Sep 30, 2009 at 7:19 AM, Ari S <sa...@gmail.com> wrote:
>>
>>>> VisualSVN Server is just properly compiled and packaged original
>>>> Subversion. So you shouldn't have difference in performance between
>>>> VisualSVN Server and CollabNet.
>>>> Could you please provide a little bit more information about your
>>>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>>>> you are used to measure peformance?
>>>
>>> Oops, I discovered that I was using the CollabNet version 1.5.4 and
>>> VisualSVN Server based on SVN 1.6.3.
>>> Could that have made such a big difference???
>>
>> I doubt it. Most changes in 1.6 were to the client layers, but even
>> setting that aside it would be rare for a server version to get
>> significantly slower.
>>
> Mark it is not true. Subversion 1.6 has significant changes in
> file-system layer which actively used on the server side:
> * Sharing multiple common representations (server)
> * FSFS repositories: Support for Memcached (server)
>
> http://subversion.tigris.org/svn_1.6_releasenotes.html#filesystem-improvements
And neither of those changes would create a 30% performance
degradation. I know there were some changes in 1.6. Just none that
would cause a serious performance drop.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402036
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ari S <sa...@gmail.com>.
Hi Mark,
> I would not expect it to hurt performance, but could you have more
> logging enabled than the default Apache? Ari have you done any kind
> of general comparison of the Apache configurations? Maybe there is
> something different in the directives used in the two configurations
> that is causing the performance difference.
No, I haven't compared the configurations yet, but I think it's a good
idea to try to do that.
The amount of logging between CollabNet and VisualSVN is probably
different, so I will examine it in more detail.
Thanks!
Ari.
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402503
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Mark Phippard <ma...@gmail.com>.
On Wed, Sep 30, 2009 at 2:15 PM, Ivan Zhakov <iv...@visualsvn.com> wrote:
> On Wed, Sep 30, 2009 at 8:39 PM, Stefan Küng <to...@gmail.com> wrote:
>> On 30.09.2009 13:31, Ivan Zhakov wrote:
>>> On Wed, Sep 30, 2009 at 3:25 PM, Mark Phippard<ma...@gmail.com> wrote:
>>>> On Wed, Sep 30, 2009 at 7:19 AM, Ari S<sa...@gmail.com> wrote:
>>>>
>>>>>> VisualSVN Server is just properly compiled and packaged original
>>>>>> Subversion. So you shouldn't have difference in performance between
>>>>>> VisualSVN Server and CollabNet.
>>>>>> Could you please provide a little bit more information about your
>>>>>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>>>>>> you are used to measure peformance?
>>>>>
>>>>> Oops, I discovered that I was using the CollabNet version 1.5.4 and
>>>>> VisualSVN Server based on SVN 1.6.3.
>>>>> Could that have made such a big difference???
>>>>
>>>> I doubt it. Most changes in 1.6 were to the client layers, but even
>>>> setting that aside it would be rare for a server version to get
>>>> significantly slower.
>>>>
>>> Mark it is not true. Subversion 1.6 has significant changes in
>>> file-system layer which actively used on the server side:
>>> * Sharing multiple common representations (server)
>>> * FSFS repositories: Support for Memcached (server)
>>>
>>> http://subversion.tigris.org/svn_1.6_releasenotes.html#filesystem-improvements
>>
>> Just a thought:
>> What Windows-SDK version were you using to compile the VisualSVN server?
>> If you used the Vista SDK (or later), then you might have triggered a
>> nasty bug in neon which can cause a massive slowdown:
>> http://svn.haxx.se/dev/archive-2008-06/0687.shtml
>> (whole thread here:
>> http://subversion.tigris.org/ds/viewMessage.do?dsMessageId=40029&dsForumId=462
>> )
>>
>>
>> Updating neon to > 0.28.2 would help.
>>
>>
> Thanks Stefan for pointing that thread. I've already seen it and of
> course we're using latest neon 0.28.6 to build VisualSVN and VisualSVN
> Server.
I would not expect it to hurt performance, but could you have more
logging enabled than the default Apache? Ari have you done any kind
of general comparison of the Apache configurations? Maybe there is
something different in the directives used in the two configurations
that is causing the performance difference.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402220
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ivan Zhakov <iv...@visualsvn.com>.
On Wed, Sep 30, 2009 at 8:39 PM, Stefan Küng <to...@gmail.com> wrote:
> On 30.09.2009 13:31, Ivan Zhakov wrote:
>> On Wed, Sep 30, 2009 at 3:25 PM, Mark Phippard<ma...@gmail.com> wrote:
>>> On Wed, Sep 30, 2009 at 7:19 AM, Ari S<sa...@gmail.com> wrote:
>>>
>>>>> VisualSVN Server is just properly compiled and packaged original
>>>>> Subversion. So you shouldn't have difference in performance between
>>>>> VisualSVN Server and CollabNet.
>>>>> Could you please provide a little bit more information about your
>>>>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>>>>> you are used to measure peformance?
>>>>
>>>> Oops, I discovered that I was using the CollabNet version 1.5.4 and
>>>> VisualSVN Server based on SVN 1.6.3.
>>>> Could that have made such a big difference???
>>>
>>> I doubt it. Most changes in 1.6 were to the client layers, but even
>>> setting that aside it would be rare for a server version to get
>>> significantly slower.
>>>
>> Mark it is not true. Subversion 1.6 has significant changes in
>> file-system layer which actively used on the server side:
>> * Sharing multiple common representations (server)
>> * FSFS repositories: Support for Memcached (server)
>>
>> http://subversion.tigris.org/svn_1.6_releasenotes.html#filesystem-improvements
>
> Just a thought:
> What Windows-SDK version were you using to compile the VisualSVN server?
> If you used the Vista SDK (or later), then you might have triggered a
> nasty bug in neon which can cause a massive slowdown:
> http://svn.haxx.se/dev/archive-2008-06/0687.shtml
> (whole thread here:
> http://subversion.tigris.org/ds/viewMessage.do?dsMessageId=40029&dsForumId=462
> )
>
>
> Updating neon to > 0.28.2 would help.
>
>
Thanks Stefan for pointing that thread. I've already seen it and of
course we're using latest neon 0.28.6 to build VisualSVN and VisualSVN
Server.
--
Ivan Zhakov
VisualSVN Team
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402217
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Stefan Küng <to...@gmail.com>.
On 30.09.2009 13:31, Ivan Zhakov wrote:
> On Wed, Sep 30, 2009 at 3:25 PM, Mark Phippard<ma...@gmail.com> wrote:
>> On Wed, Sep 30, 2009 at 7:19 AM, Ari S<sa...@gmail.com> wrote:
>>
>>>> VisualSVN Server is just properly compiled and packaged original
>>>> Subversion. So you shouldn't have difference in performance between
>>>> VisualSVN Server and CollabNet.
>>>> Could you please provide a little bit more information about your
>>>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>>>> you are used to measure peformance?
>>>
>>> Oops, I discovered that I was using the CollabNet version 1.5.4 and
>>> VisualSVN Server based on SVN 1.6.3.
>>> Could that have made such a big difference???
>>
>> I doubt it. Most changes in 1.6 were to the client layers, but even
>> setting that aside it would be rare for a server version to get
>> significantly slower.
>>
> Mark it is not true. Subversion 1.6 has significant changes in
> file-system layer which actively used on the server side:
> * Sharing multiple common representations (server)
> * FSFS repositories: Support for Memcached (server)
>
> http://subversion.tigris.org/svn_1.6_releasenotes.html#filesystem-improvements
Just a thought:
What Windows-SDK version were you using to compile the VisualSVN server?
If you used the Vista SDK (or later), then you might have triggered a
nasty bug in neon which can cause a massive slowdown:
http://svn.haxx.se/dev/archive-2008-06/0687.shtml
(whole thread here:
http://subversion.tigris.org/ds/viewMessage.do?dsMessageId=40029&dsForumId=462
)
Updating neon to > 0.28.2 would help.
Stefan
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402170
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ivan Zhakov <iv...@visualsvn.com>.
On Wed, Sep 30, 2009 at 3:25 PM, Mark Phippard <ma...@gmail.com> wrote:
> On Wed, Sep 30, 2009 at 7:19 AM, Ari S <sa...@gmail.com> wrote:
>
>>> VisualSVN Server is just properly compiled and packaged original
>>> Subversion. So you shouldn't have difference in performance between
>>> VisualSVN Server and CollabNet.
>>> Could you please provide a little bit more information about your
>>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>>> you are used to measure peformance?
>>
>> Oops, I discovered that I was using the CollabNet version 1.5.4 and
>> VisualSVN Server based on SVN 1.6.3.
>> Could that have made such a big difference???
>
> I doubt it. Most changes in 1.6 were to the client layers, but even
> setting that aside it would be rare for a server version to get
> significantly slower.
>
Mark it is not true. Subversion 1.6 has significant changes in
file-system layer which actively used on the server side:
* Sharing multiple common representations (server)
* FSFS repositories: Support for Memcached (server)
http://subversion.tigris.org/svn_1.6_releasenotes.html#filesystem-improvements
--
Ivan Zhakov
VisualSVN Team
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402018
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Mark Phippard <ma...@gmail.com>.
On Wed, Sep 30, 2009 at 7:19 AM, Ari S <sa...@gmail.com> wrote:
>> VisualSVN Server is just properly compiled and packaged original
>> Subversion. So you shouldn't have difference in performance between
>> VisualSVN Server and CollabNet.
>> Could you please provide a little bit more information about your
>> configuration: server os, client os, IPv4 or IPv4, etc. What operation
>> you are used to measure peformance?
>
> Oops, I discovered that I was using the CollabNet version 1.5.4 and
> VisualSVN Server based on SVN 1.6.3.
> Could that have made such a big difference???
I doubt it. Most changes in 1.6 were to the client layers, but even
setting that aside it would be rare for a server version to get
significantly slower.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402016
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ari S <sa...@gmail.com>.
Hi Matthieu,
> Probably asking the obvious, but do you have the same repository
> format on both?
Yes, they have the same format; they are actually exact copies of the
same repository converted from CVS using the 'cvs2svn' script /
program.
> Something you can try (if it's not already the case) is to run the two
> servers on the exact same repositories (i.e. not two copies of the
> same, but really the same thing in the filesystem).
I initially didn't do that, because I feared one server would clobber
the repository for the other.
But I guess now it's okay, esp. for read-only tests.
Thanks for the idea!
Best regards,
Ari.
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402502
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Matthieu Moy <Ma...@grenoble-inp.fr>.
Ari S <sa...@gmail.com> writes:
> On the server, we installed the CollabNet server 1.5.4 on port 80, and
> the VisualSVN on port 8080.
> Both use standard Apache authentication (not Windows).
Probably asking the obvious, but do you have the same repository
format on both?
Something you can try (if it's not already the case) is to run the two
servers on the exact same repositories (i.e. not two copies of the
same, but really the same thing in the filesystem).
--
Matthieu Moy
http://www-verimag.imag.fr/~moy/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402057
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ari S <sa...@gmail.com>.
Hi Ivan (and Mark),
> VisualSVN Server is just properly compiled and packaged original
> Subversion. So you shouldn't have difference in performance between
> VisualSVN Server and CollabNet.
> Could you please provide a little bit more information about your
> configuration: server os, client os, IPv4 or IPv4, etc. What operation
> you are used to measure peformance?
Oops, I discovered that I was using the CollabNet version 1.5.4 and
VisualSVN Server based on SVN 1.6.3.
Could that have made such a big difference???
The server is a Pentium-4 2.8 GHz with 2 Gb of RAM, and a 500 + 320 Gb
hard drives.
The clients have the same processor and amount of RAM.
Both the server and the clients run Windows XP-SP 3.
We use IPV4 on a local network.
On the server, we installed the CollabNet server 1.5.4 on port 80, and
the VisualSVN on port 8080.
Both use standard Apache authentication (not Windows).
What we did to compare the performance is a clean check-out of a 143
Mb repository.
As clients we used:
- cmd line client 1.5.4
- Tortoise SVN 1.6.5
So I guess the question now is: has anyone noticed a 30% performance
difference between Subversion 1.5.4 and 1.6.3??
(I tried upgrading the CollabNet 1.5.4 version to 1.6.3 but for some
strange reason failed on the current test machine.
Will try to reinstall both servers on a clean machine)
Best regards
Ari.
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2402012
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ivan Zhakov <iv...@visualsvn.com>.
On Tue, Sep 29, 2009 at 7:22 PM, Ari S <sa...@gmail.com> wrote:
> Hi Mark,
>
>> I would assume that VisualSVN is largely the same (does not customize
>> its binaries). I know they toss in some Windows-specific tools and
>> configuration UI. My guess is that VisualSVN is configuring Apache
>> for Windows authentication or something like that. So to do a real
>> comparison, you would need to configure the Apache server provided in
>> the CollabNet distribution to do something similar. Or conversely, if
>> the VisualSVN server allows you to use plain Apache authentication
>> then you could try it that way.
>
> Thanks for your reply!
> I did set up VisualSVN _and_ CollabNet such that they both use Apache
> authentication.
> Without that the difference was even 100% (VisualSVN twice as slow as
> Collabnet).
>
VisualSVN Server is just properly compiled and packaged original
Subversion. So you shouldn't have difference in performance between
VisualSVN Server and CollabNet.
Could you please provide a little bit more information about your
configuration: server os, client os, IPv4 or IPv4, etc. What operation
you are used to measure peformance?
--
Ivan Zhakov
VisualSVN Team
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2401707
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Ari S <sa...@gmail.com>.
Hi Mark,
> I would assume that VisualSVN is largely the same (does not customize
> its binaries). I know they toss in some Windows-specific tools and
> configuration UI. My guess is that VisualSVN is configuring Apache
> for Windows authentication or something like that. So to do a real
> comparison, you would need to configure the Apache server provided in
> the CollabNet distribution to do something similar. Or conversely, if
> the VisualSVN server allows you to use plain Apache authentication
> then you could try it that way.
Thanks for your reply!
I did set up VisualSVN _and_ CollabNet such that they both use Apache
authentication.
Without that the difference was even 100% (VisualSVN twice as slow as
Collabnet).
Thanks for the additional information on what goes into your packages.
Best Regards,
Ari
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2401668
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].
Re: Comparison Collabnet and VisualSVN
Posted by Mark Phippard <ma...@gmail.com>.
On Tue, Sep 29, 2009 at 9:53 AM, Ari S <sa...@gmail.com> wrote:
> Currently, we are investigating a migration from CVS to Subversion.
> For the Subversion server, we have compared specs of different
> suppliers, and the shortlist has now been narrowed down to two:
> CollabNet and VisualSVN Server.
>
> When performing performance tests with the two, we found CollabNet to
> be about 30% faster than VisualSVN server when using "out of the box
> configurations" (as little changed as possible).
>
> Has anyone else found similar (or totally different) figures?
>
> Are there other pros and cons that should be considered in choosing
> between these two vendors?
>
> Could anyone perhaps point me to such performance (or feature)
> comparison investigations done in the past by others?
I manage the CollabNet binaries. There is no particular reason our
binaries should be any faster or slower than other SVN binaries. We
are basically providing packaging and certification. We compile
Apache 2.2.x and Subversion and dependencies etc. and provide them
with a Windows installer. Before we release the packages we put them
through an internal qualification process, which is largely focused
around the packaging and installation process. There are no tweaks or
customizations being done on the binaries themselves, other than the
normal compile time options that the various products allow. We also
include ViewVC in our packaging.
I would assume that VisualSVN is largely the same (does not customize
its binaries). I know they toss in some Windows-specific tools and
configuration UI. My guess is that VisualSVN is configuring Apache
for Windows authentication or something like that. So to do a real
comparison, you would need to configure the Apache server provided in
the CollabNet distribution to do something similar. Or conversely, if
the VisualSVN server allows you to use plain Apache authentication
then you could try it that way.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2401643
To unsubscribe from this discussion, e-mail: [users-unsubscribe@subversion.tigris.org].