[THIN] XenApp 5 - 2008 x64 Slow
- From: Andy Friar <Andy.Friar@xxxxxxxxxxx>
- To: "thin@xxxxxxxxxxxxx" <thin@xxxxxxxxxxxxx>
- Date: Thu, 6 Jan 2011 15:10:02 +0000
I'm have a problem with the following Windows 2008 SP2 Enterprise x64
XenApp 5 Enterprise R01
Boxes are Dell blade 8 core + HT 49Gb RAM
Local manadatory profiles with full Appsense suite utilising EM +
Personalisation service.
Desktop / My Documents / Favourites redirected to a Netapp Filer running CIFS.
The more users that come onto the box the slower the machine gets.
@ 0-45 users the box is responsive, logon times are good but it still has this
apparent lag when accessing local apps
@ 50 the lag is getting worse sometimes freezes but always comes back.
@ 60 logon times are substantial, just selecting a cell in excel can take 30+
seconds.
Performance counters seem fine
14% processor time
25k context switches / sec no massive jumps
0 Processor queue length
99% disk idle
0 disk queue length
15-20k threads
0 Redirector\Current Commands
The only large one is system calls / sec which can be anywhere from 80k to
1.5m, I've cut down the apps but system calls are still high.
To me the server feels as if it has no cpu left, but given the resources free
it looks more like an i/o problem, I've tuned the smb with the following
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Lanmanserver\Parameters]
"MaxWorkItems"=dword:00002000
"MaxMpxCt"=dword:00000800 (subsequently has
been changed to match the Netapp)
"MaxRawWorkItems"=dword:00000200
"MaxFreeConnections"=dword:00000064
"MinFreeConnections"=dword:00000020
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Lanmanworkstation\Parameters]
"MaxCmds"=dword:00000800
[HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\Session
Manager\Configuration Manager]
"RegistryLazyFlushInterval"=dword:0000003C
The NetApp filer has had cifs.max_mpx set to 1124 so I've subsequently changed
the citrix servers to match this number rather than the 2048 previously.
It's running around 30%cpu 2000 io/sec 98% cache hit.
Any suggestions on where else I might look into what is causing the bottleneck?
Thanks
Andy
Other related posts:
