Can Samba scale with NT_Terminal_Server? (was: Samba and WinTS / WinDD)

[Charles N. Owens]

I'm wondering if the unique (?) way NT Terminal Server (NT_TS) connects
to Samba is a problem.  Samba seems to spawn a separate process per
connecting client.  With "normal" single-user stations (Win3.x, Win9x,
WinNT), this results in a process per user.  But with NT_TS (and other
related Citrix derivatives) we have many users coming from a single
client machine, all of which get handled by a _single_ Samba process
(one per NT_TS server box).  It is not unreasonable to expect that there
may then be 50 to 150 users being handled by a single Samba process
(depending on the horsepower of the NT_TS server).

This difference (many users per Samba process vs. one) seems like it
could be a big deal to me.  Is Samba known to be able to properly handle
the many-user case with as many users as I describe above?  Was it at
least _designed_ to be able to do so?

A separate question is whether or not NT_TS talks to a PDC any
differently than with plain NT.  Something is making our NT_TS boxes
unstable (beyond what's expected given their pedigree ;-).  It seems to
be related to their being involved with a Samba PDC or with their trying
to relatively heavy file sharing via Samba (or possibly both).

I'd be most appreciative if anyone had any experiences to share to
suggest the best way to proceed.  What's the best way to trouble-shoot
this?

We've been running multiple pilot installations of NT_TS + Citrix
Metaframe [v1.8] with varying results.  Here's what we have:

[In all cases user apps are Netscape 4.61, Office 97, Eudora 4.x,
installed on local NT_TS drives.  All servers have Service Pack 4 plus a
number of fixes we've squeezed out of Microsoft and Citrix.]


A:  	NT_TS + Metaframe server in workgroup mode.
	File, WINS service: Samba 1.9.18(?) box.
	Load:  primary computing environment for about 6 users
	Result:  Rather stable.  Will eventually blue-screen if
		up for more than a week or three.

B:	NT_TS + Metaframe in Samba PDC domain
	PDC, WINS, File+Print:  Samba -HEAD branch, circa mid-July
	Load:  30 users (using "Windows Terminals" [Neoware])
	Result:  Unstable.  Tends to blue-screen on average every
		1.5 days.  Rebooting every morning seems to keep
		things livable.

C:	NT_TS + Metaframe in Samba PDC domain (not the same as B)
	PDC, WINS:  Samba -HEAD branch from August 11th
	File/Print:  Samba 2.0.5a
	Load:  10 users.  Five are using lousy 16-bit app that uses
		SMB-file share (samba) as shared-access database.
	Result:  Unstable.  Similar to B.

With B & C I've seen funky file behavior.  Things work fine most of the
time, but after a day or two of uptime NT_TS may start acting sluggish
(screen freezes here and there, sessions dropping)... folks may not be
able to save their work (as seen by Matt J. in his message), often being
told that the disk is full.  At some point the system will usually
blue-screen... it comes down hard... it isn't able to save a crash dump
or auto-reboot as we've told it (despite our extensive coaxing efforts).

The event log will generally have a number of these message right before
the time of the crash:

	The redirector received an SMB that was too short.

It may also contain one or more of these:

	The redirector has timed out a request to PDC-ENC-DS.
		[Samba PDC for example C]
	The redirector has timed out a request to smb1.
		[Samba File/print box for example C]
	The redirector received an incorrectly formatted response
             from smb1. 

I'd hoped that some of my trouble with example B was due to trying to do
my file/print serving from Samba-HEAD.  Thus the switch to 2.0.5a for
these services in example C.  Doesn't seem to have helped much, making
me wonder if the major trouble is lies with some PDC interaction.  [
perhaps it's time to upgrade my -HEAD branch PDC code? ]

It would be best, I'm sure if I did an implementation of this with a
true NT PDC and Samba for just file/print.  Then I could focus just on
file issues.  Or visa versa so I could focus on Samba PDC issues.  I've
definitely muddied the waters with my current approach.  (sigh)

So... any other NT_TS users doing the Samba thing with better luck than
I?

And, of course, I feel duty-bound to point out that were NT_TS and
Metaframe correctly, or at least more robustly implemented, they'd be
able to with-stand non-ideal behavior from Samba (assuming this is
occurring) without going belly-up.  Said another way, it's not Samba's
fault that there are some fundamental flaws in NT_TS that cause it to
crash when tickled.  Samba is only responsible for the tickling.

[ The same argument applies to behavior I've seen with the ICA client
software when accessing ICA-mapped local drives.  If you bang hard
enough, you'll crash the server (though a Citrix hotfix mostly cleaned
this up... now it just kills your session ;-) ]

But I digress...

Thanks,

Charles O.


Matthew Jamison <xmj at cypress.com> said:

> I have noticed a small problem with WinTS/WinDD and samba. If for any 
> reason the samba server goes down and then comes back up while some one is 
> editing a MS document on one of these boxes the user is unable to save there 
> work. I think this is strange since this does not seem to effect NT 
> workstation and Win9x. 


-- 
-------------------------------------------------------------------------
  Charles N. Owens                               Email: owensc at enc.edu
                                            http://www.enc.edu/~owensc
  Network & Systems Administrator
  Information Technology Services  "Outside of a dog, a book is a man's
  Eastern Nazarene College         best friend.  Inside of a dog it's 
                                   too dark to read." - Groucho Marx
-------------------------------------------------------------------------