Hello,

I believe that the issue you are describing is indeed caused by problems
with the network storage. I would typically suggest that you avoid using
remote disks for repository storage when possible. Local disks are the best
option to store repositories in terms of performance and server response
time. Network storage can add a significant performance penalty, so *use
*local* SSD storage for your repositories (and working copies as well)*.

>  1. Can FSFS repositories hosted on CIFS/SMB network storage experience
significant performance degradation due to storage latency or locking
behaviour?
>  4. Are there any known performance or reliability concerns with hosting
production FSFS repositories on CIFS/SMB shares?

Network storage can add a significant performance penalty on its own and
adds an additional layer where things can go wrong. If the storage or
network has problems (network-related or load related, for example), this
can make things much worse. Any software that uses the same network storage
can be affected, not just SVN. You mention Unreal cache reimports from the
same network storage have become slower too. I don't think that these
caches use SVN at all (please correct me if I'm mistaken here), so this
again points to problems with the network storage.

>  2. Are the observed Apache D-state processes and CIFS-related wait
channels consistent with delays accessing repository storage?

As far as I can tell, D state means "uninterruptible sleep"
<https://man7.org/linux/man-pages/man1/ps.1.html#PROCESS_STATE_CODES>,
which usually indicates the the process is waiting for I/O:
[[[

               D    uninterruptible sleep (usually I/O)

]]]

The channels specified in your email points at CIFS: *cifs_*call_async,
*cifs_*readdata. So it seems that the process is waiting for your network
share. So, again, the network storage is IMO the most likely culprit here.

You've provided a lot of details (thanks for that!), but a few things are
still missing in your report. Which versions of SVN and httpd are you
running on the server? What versions of APR and APR-Util are they using? I
recall that by default CollabNet SVN Edge was using TCP 18080
<https://stackoverflow.com/a/35059641>, which you mentioned in your email.
I think that SVN Edge was using SVN 1.8.x, which isn't supported for years.
The latest version is SVN 1.15.

The output of *svn --version --verbose *ran on the server against the
binaries of your SVN server installation can help identify versions of the
components of your SVN server.

BTW, are there any errors or warnings in the httpd logs
<https://svnbook.red-bean.com/nightly/en/svn.serverconfig.httpd.html#svn.serverconfig.httpd.extra.logging>
?

Thank you.

On Wed, 18 Nov 2026 at 15:54, Ananya Srivastava <
[email protected]> wrote:

> Hello,
>
> I'm looking for guidance from Subversion administrators regarding a
> performance issue in our production SVN environment.
>
> We have been experiencing significant performance degradation across
> multiple Unreal Engine projects. SVN updates, commits, checkouts and even
> opening the TortoiseSVN Repository Browser have become slow.
>
> I'm trying to identify whether the issue relates to Subversion, Apache
> HTTPD, repository storage, CIFS/SMB or the underlying infrastructure.
> 1. Our environment
>
>    -
>
>    Operating System: Ubuntu 20.04.5 LTS
>    -
>
>    Kernel: Linux 5.15
>    -
>
>    Subversion access: Apache HTTPD over HTTPS
>    -
>
>    SVN port: 18080
>    -
>
>    Repository format: FSFS
>    -
>
>    Repository storage: CIFS/SMB network share
>    -
>
>    CIFS mount: //moon/repo mounted at /mnt/repo
>    -
>
>    SMB version: 3.1.1
>
> The repositories are stored under:
>
> /mnt/repo/data/repositories/
> 2. Performance symptoms
>
> We are experiencing the following:
>
>    -
>
>    SVN updates and commits have become significantly slower across
>    multiple projects.
>    -
>
>    TortoiseSVN Repository Browser took approximately 9 minutes to
>    initialize.
>    -
>
>    A 14.7 GB checkout took approximately 13 minutes.
>    -
>
>    A 2.7 GB commit took approximately 4 minutes.
>    -
>
>    A previous 15 GB commit completed in approximately 5–7 minutes.
>    -
>
>    Unreal Engine cache reimports from network storage have also become
>    slower.
>    -
>
>    Unreal rendering performance on the farm has reportedly degraded as
>    well.
>
> Artists use local Unreal working copies, while central SVN repositories
> are stored on network storage.
> 3. Diagnostics performed
>
> During a slow 14.7 GB checkout, we observed:
>
>    -
>
>    Linux I/O pressure reaching approximately 23%.
>    -
>
>    Multiple Apache httpd processes in uninterruptible D state.
>    -
>
>    Kernel wait channels included:
>    -
>
>       cifs_call_async
>       -
>
>       compound_send_recv
>       -
>
>       cifs_readdata
>       -
>
>       wait_on_page_bit_common
>
> The SVN repository storage is mounted through CIFS/SMB.
>
> TCP connectivity to the SVN server on port 18080 is successful.
> 4. Questions
>
>    1.
>
>    Can FSFS repositories hosted on CIFS/SMB network storage experience
>    significant performance degradation due to storage latency or locking
>    behaviour?
>    2.
>
>    Are the observed Apache D-state processes and CIFS-related wait
>    channels consistent with delays accessing repository storage?
>    3.
>
>    What is the recommended method to distinguish between Subversion
>    processing delays, Apache HTTPD delays, CIFS/SMB latency and physical
>    storage bottlenecks?
>    4.
>
>    Are there any known performance or reliability concerns with hosting
>    production FSFS repositories on CIFS/SMB shares?
>    5.
>
>    What read-only diagnostic commands or measurements would you recommend
>    before making any configuration changes?
>    6.
>
>    Are there any particular Apache HTTPD, mod_dav_svn or Subversion
>    settings that we should investigate?
>
> We are not looking to make random configuration changes or rebuild
> repositories without understanding the root cause.
>
> Any guidance on systematically identifying the bottleneck would be greatly
> appreciated.
>
> *We have also involved our internal IT and production teams in parallel to
> investigate the infrastructure and application-side components.  *
>
> Thank you.
>
>
> ____________________________________________________________________________
> Disclaimer: - This email and any files transmitted with it are strictly
> confidential and intended solely for the use of the individual or entity to
> whom they are addressed. If you have received this email in error, please
> notify the system manager. This message contains confidential information
> and is intended only for the individual named. If you are not the named
> addressee, you should not disseminate, distribute or copy this email.
> Please notify the sender immediately by email if you have received this
> email by mistake and delete this email from your system. If you are not the
> intended recipient, you are notified that disclosing, copying, distributing
> or taking any action in reliance on the contents of this information is
> strictly prohibited.
>


-- 
With best regards,
Pavel Lyalyakin
VisualSVN Team

Reply via email to