[THIN] Re: TSCALE or Appsense
- From: "Rick Mack" <ulrich.mack@xxxxxxxxx>
- To: thin@xxxxxxxxxxxxx
- Date: Fri, 12 Jan 2007 21:09:52 +1000
Hi Jeremy,
You're welcome to a copy of the template. In a normal scenario, maxworkitems
at 4K at a school it might be 64K.
SMB tuning aside, the other really useful thing it can be used to do is to
disable RDP client printer mapping. Otherwise administrators can install a
whole bunch of foreign printer drivers on their servers without even
noticing.
If you want to see what the maxmpxct tuning does in a really dramatic way,
you have to get something like Corel Draw 9.
Example 1.
In this particular real example all the servers (Citrix and file/print) were
top end IBM servers and the SAN was IBM as well ;-)
Start with an untuned file server.
Corel Draw 9 has a 14K coreldrw.ini file which it reads in 1 byte chunks. So
if coreldrw.ini was on the user's home drive it generates 14,000 separate
network i/o requests on startup. Now put the application in a school where
you've got a classroom of kids starting the app at the same time and stand
back. Note that the students are generally spread over as many as 8-10
Citrix servers.
Suddenly your whole Citrix farm will hang, but if you wait long enough the
Citrix servers will recover after 15-20 minutes, generally about 5 minutes
longer than you're willing to wait before you reboot them.
If you close all the shared folder sessions on the file server the Citrix
Servers will recover instantly.
If you max out mpxct/maxworkitems and reboot and repeat the "test", things
slow down for a bit but nothing hangs.
Peter Ghostine found the same thing actually goes for Citrix and mapped
client drives. If you've got a lot of client drive traffic, it can
cause hangs because now the TS system is a file server, essentially to
itself. Bump up maxmpxct etc on the TS/Citrix system and the hangs go away.
That's why my TS server tuning adm template includes the same maxmpxct etc
section as the back-end server adm template.
Otherwise, monitor redirector current commands and work out how much free
queue space you've got with the default queue size. Then think about what
would happen if the file server got busy and the free queue size dropped
even more.
Example 2.
A farm of about 20-30 Citrix servers, about 1200-1400 concurrent users.
Throw in an AD site location bug so that the server farm is using a DC
across a 2 MB WAN link.
In the mornings the servers were hanging for between1 minutes to 5 minutes
at times while users were logging in in the mornings. After logon things
were fine.
Increased MaxMpxCT/maxworkitems on the DC and the hangs went away. Fixed the
site location bug too, but the hangs were "fixed" before that.
SMB tuning (maxmpxct/maxworkitems) may not make things work faster when
everything is humming along, but without it you can have a very
uncomfortable time when things get busy.
Remember that in the first example we were talking about 30 x 14K of ini
file data (30+ MB of network traffic), but about 500,000 separate network
i/o requests within a few seconds. That isn't a whole heap of data, but it
sure did make the file server busy, and the Citrix servers just froze.
I rest my case.
--
Ulrich Mack
Commander Australia
Other related posts:
