Recently made the decision to setup a new RDP farm based on Server 2016, I'd better went for Server 2012R2....
Our infrastructure
We have one Remote Desktop Connection Broker server, and 2 Remote Desktop Session Hosts, hosting the users.
The users have User Profile Disks being hosted on a 3rd server (a File & Print server).
2 physical servers hosting each a virtual RDP server with 26 vCPU's and 50G memory.
The file and print server, domain controller, a 3rd party mail server and SQL server are as well hosted on the physical hosts. No CPU sharing among different VM's.
Each physical host is equipped with :
CPU : Dual Intel(R) Xeon(R) CPU E5-2630 v4 (dual 10 core)
Memory : 196G
Disk : Dell MD3xxx storage (IO is not an issue, more than 150 megabyte per second is possible, same with latency, no issues there)
Each RDP server hosts around 20 to 25 users.
Our issues :
RDP users complain about a very bad user experience. This behaviour is seen on a wide level. All users experience the issue.
Users are sitting on different locations, so the network towards the RDP servers is not causing the issue(s). The location where the most users are connecting from it's performance is good -> Success rate is 100 percent (100/100), round-trip min/avg/max
= 12/15/24 ms.
In general, a high CPU usage is noticed as well. For example doing a speedtest in a browser, pulls more than 20 to 30% CPU (total CPU usage on the 26 cores...).
To be a bit more specific ;
- Outlook (and actually office) works slow in general. Scrolling, browsing through pages, ...
- Icons on the desktop are flickering when the CPU raises to 70 to 80%
- Browsing through windows, minimizing, maximising, ... goes slow.
- Logging in goes slow, takes around 1 or 2 minutes to logon. It starts with a black screen, followed by the task bar, the desktop items, ...
- Opening the start menu, rightmouseclick on different items goes with a delay
- Browsing through a specific news page hangs the browser, en eventually the complete session (for example opening Internet Explorer with the msn.com startpage, causes the browser to "hang" for a number of seconds).
What we've already tried :
- Adding a massive amount of resources (26 vCPU's including 50G memory), even with 1 user, no fluent session feeling.
- Using the VMWare optimization tool. No difference experienced. Used the Windows 2016 RDSH template
- Played with TCP Offload disable/enable settings, no results.
- Doing file transfers from and to the File & print server, they go well.
- Disabling and enabling fair share for network, drives & CPU : no difference
We have the feeling something ... is stopping the server of being fast and quick. But the question is what?
Out of options...
Thanks in advance for your help!
J