samba - Sep 2006 - sizing a print server for 10'000 users

Hi,

I'm building a print-to-PDF queue system for a customer. A Samba queue pass
jobs to Ghostscript. The PDF output is then provided back to the user (by email
or web URL).

My question is on server sizing. It will be on Linux (SLES 9) and will only host
the print queue, nothing else. Samba will be joined in the ADS domain, so I can
recognize the submitting user (%u).

Every Windows client will have the print queue defined. What happen with the
connection ? do I assume Samba will see 10'000 permanent connections ? Or
only when printing ?

And then, when several client submit print jobs at the same time, does Samba
serialize the calls to my script ? Or should I assume I can be called
10'000x in parallel (assuming 10'000 users click "Print" at
the same time) ?

I'm as well interested at building a small "admin" web page where
one can see the queue, the last 10 job entries, the load, possibly a usage
graph, etc.

Hints from a similar setup are welcome. 
TIA,
Charles

-- 
Charles Bueche <charles@bueche.ch>
sand, snow, wave, wind and net -surfer

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/13/2006 11:18 AM, Charles Bueche escreveu:
> Hi,
> 
> I'm building a print-to-PDF queue system for a customer. A 
> Samba queue pass jobs to Ghostscript. The PDF output is
> then provided back to the user (by email or web URL).
	Do you really need to use samba for that? You can
have CUPS working as a print-to-PDF system. I don't have
anything against Samba, but you could reduce one layer in
your design.

> My question is on server sizing. It will be on Linux 
> (SLES 9) and will only host the print queue, nothing
> else. Samba will be joined in the ADS domain, so I can
> recognize the submitting user (%u).
	Hmmm, ok, it answer partially the above question,
perhaps you can have the same behaviour using IPP.

> Every Windows client will have the print queue defined. 
> What happen with the connection ? do I assume Samba
> will see 10'000 permanent connections ? Or only when
> printing ?
	AFAIR, printer shares don't stay "open", which
means that the share will be _really_ used when you
print to it.

> And then, when several client submit print jobs at the 
> same time, does Samba serialize the calls to my script ?
>  Or should I assume I can be called 10'000x in parallel
> (assuming 10'000 users click "Print" at the same time) ?
	I think you will have bottlenecks in Samba and
Ghostscript itself, because you need time to generate
the PDFs and you will also have the bottleneck of write
to disk and send e-mails.

	Depends on the implementation of your script,
but you can easily figure that out testing with only
two clients and big files, I don't think your problem
will be related to the number of clients but with the
size of files.

> I'm as well interested at building a small "admin" web 
> page where one can see the queue, the last 10 job
> entries, the load, possibly a usage graph, etc.
	Sounds like CUPS to me. :)

> Hints from a similar setup are welcome. 
> TIA,
> Charles
	I think you can extend CUPS to what you need.

	Kind regards,

- --
Felipe Augusto van de Wiel <felipe@paranacidade.org.br>
Coordenadoria de Tecnologia da Informa??o (CTI) - SEDU/PARANACIDADE
http://www.paranacidade.org.br/           Phone: (+55 41 3350 3300)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)
Comment: Using GnuPG with Debian - http://enigmail.mozdev.org

iD8DBQFFCrjECj65ZxU4gPQRApwBAJ90q58Ly6Okl0djO1uE9JkiqoPjSgCgvmLb
+x7MMVzGKHL5CxdOauQNSfA=8wGV
-----END PGP SIGNATURE-----