We are experiencing performance issue on Windows Server 2012 R2 Standard with the RDS role installed
after migrating from Windows Server 2008 R2.
It happens from time to time in peak hours. It looks like users who are already on the server do not notice great degradation in
performance, but there is a big problem with logging on the server by new users - it is not possible at that time. There are no error messages, the session starts, but then the screen is usually black. For the existing users, that are
already logged in, they can work with the already open programs and resources, but no new programs or resources can be open\used.
The problem occurs in peak hours when over 200 sessions per host are reached (sometimes problem occurs
about 100 sessions) Before the upgrade serves could serve over 300 sessions.
To load balance users we use “Network Load balancing”. In our environment we have 5 servers in nlb cluster (IGMP Multicast). On
4 server we have: Remote Desktop Session Host, Remote Desktop Connection Broker and Remote Desktop Web Access on one of the servers just Remote Desktop Session Host (we haven’t notices any difference between these servers regarding performance). There
are two domain controlers. Profiles and users' files are loaded from two separate file servers.
Observations:
- When we left just one serve it was able to serve 320 users without problem.
- Another observation is when the problem with logon starts there is no problem to logon on local user through ibm management. The same user is not able to logon through rds at the
same time.
- In the same subnet and in the same domain we have second nlb cluster based on Windows Server 2008 R2 and there are no problems.
- On our RDS servers process svchost.exe associated with TermService takes over 5 GB of RAM is it a proper value?
- When problem with slow logging occurs there are usually also problems with opening "Event Viewer" --> "RPC is too busy to complete this operation"
- WMI c:\windows\system32\wbem\repository\OBJECTS.DATA (from time to time exceeds 1 GB)
Hardware parameters of the server
IBM x3550 M3 [794482G]
CPU: 2 x Xeon X5690 3.47 GHz
RAM: 96 GB
NIC: Broadcom BCM5709C NetXtreme II (updated to version 7.12.3.0)
What have done already and did not help:
- TCP Chimney is on default (auto) turning off doesn't help.
- changing Maximum Number of RSS Queues on network card from 2 (default) to 8
- mapping just default printer from the client
- turning off desktop experience
- turning off UDP on RDP sessions
- turning off smartcard service\nMPN ID:528137
- server is fully up to date (including KB 2887595)
- In group policy setting: "Turn off Automatic Root Certificates Update" to Enable.
- AV exclusions based on https://support.microsoft.com/en-us/kb/822158 and http://blogs.technet.com/b/askperf/archive/2014/08/08/wmi-repository-corruption-or-not.aspx
- Increased WMI MemoryPerHost (__providerhostquotaconfiguration) to 1GB
from 512 MB. (http://blogs.technet.com/b/askperf/archive/2008/09/16/memory-and-handle-quotas-in-the-wmi-provider-service.aspx and http://blogs.technet.com/b/configmgrteam/archive/2009/05/08/wmi-troubleshooting-tips.aspx)
Workarounds:
- Restart the server
- Sometimes helps rebuilding WMI
Anybody has similar experiences? Any solutions?