We are migrating our On-Prem RDS infrastructure to Azure.
150 users, no Remote Apps, all straight desktops. Office 365, a Java based 5250 emulator and a IE11 is pretty much all we run.
On-Prem is 1 Broker and 6 session host (Hyper-V, Server 2012-R2, 12vCPU, 48GB RAM). UPD on an iSCSI SSD NAS, and redirected folders to a regular file server on our 8GB Fiber SAN. I average 25-30 users per host on prem and do not have performance issues.
In Azure we are testing with 40 users. 4 session host (Server 2016, 8vCPU, 32GB RAM, P30 premium SSD) UPD's and folder redirection are to a file server in Azure with P30 SSD. When load testing with 40 users logging in at roughly same time, we see horrible performance. 10-20+ minute logins, servers lock up (not CPU or memory constrained). I can get about 4-6 users connected then performance craters. It cannot be the app load, as we are not even getting to the desktops.
We have added the patch and registry entries to not create the firewall rules for the servers. We have changed from UPD to roaming profiles. We have opened a SEV1 ticket with MS (via our cloud MSP). In the end, the MSP is saying it is the RDS config, and that they cannot help us.
I believe the issue is the throughput limitation for the VM's in Azure. We ran PerInsight on one of the VM's during a test. The disk latency was 5 seconds. Not 5 milliseconds, 5 seconds.
For those of you running RDSH in Azure, what VM sizes are you using, and what type of user/session host density are you getting? Are you using UPD's with folder redirection or Roaming profiles? What are the best practices or Gotcha's I need to be concerned about or aware of?
Thanks for reading, and appreciate any insight you may have.
James