Windows<->VMS Samba performance issue

John E. Malmberg wb8tyw at qsl.net
Fri Sep 17 19:04:45 GMT 2004


In article <1095443965.17913.22.camel at bgpc.dymaxion.ca>,
    BG - Ben Armstrong <BArmstrong at dymaxion.ca> writes:

> We have observed that a Linux 2.6 client using smbfs completes a simple
> "write large files" benchmark of our own devising* 10 times faster than
> a Windows client on the same network with the same load!  This suggests
> that there is great room for improvement just by changing some settings
> -- possibly some of the socket options, as a casual review of the Samba
> literature suggests.

VMS version?
TCPIP platform and version?
Samba version?

Any work in that area would be appreciated.  Also if you can run the
torture tests against the SAMBA 2.2.8 to indicate any functionality that is
not working properly.  The tests supplied with the SAMBA kit do not seem
to have been ported to OpenVMS.

There are a number of tuning issues that I addressed in a SAMBA VMS FAQ,
of which the latest copy I did is available from a GOOGLE search, even
though it is quite old.

In general, some of the socket options may help.  What I would expect to
make the biggest difference is setting the RMS default buffer sizes.

There is one key difference in how some Microsoft Windows clients treat
large files and how SAMBA does.  SAMBA sends the file sequentially from
start to finish.  For some unknown reason, Microsoft Windows sends the
first part of the file, skips a bit and sends the middle, and then backfills.

This may affect how OpenVMS is handling performance.

And it just may be a case where SAMBA to SAMBA transfers are more efficient
than WINDOWS to SAMBA transfers, since you would expect that if it were
really a server issue, that the server would perform badly for both clients.

If you use a LINUX SAMBA server instead of a VMS SAMBA server, do you see
the same difference in performance?

The SMBD.EXE image needs to be installed as shared to reduce the virtual
memory load and to speed up the startup of new instances.  This is not likely
to help the large file transfer much, but it is a related to do option.

> So what we're after now is some pointers for tuning so that we can
> realize with a Windows client a similar level of performance to the
> Linux client.  We need to know how to find out which settings are out of
> whack, and how to choose from among the thousands of possible knobs we
> could fiddle with.

General VMS tuning should be looked at.  Make sure that WSMAX / WSEXTENT is
large enough to keep processes from paging excessively.  The RMS buffering
mentioned above.

You also will want to make sure that your non-page pool is adequate.  With
your normal steady state load, the usual recomendation is that there should
be 300,000 bytes free, with out any pool expansion having taken place.

Or were you looking for Windows client tuning tips?

-John
wb8tyw at qsl.net
Personal Opinion Only



More information about the samba-vms mailing list