Dear Everybody ! The Windows 2019 server responds very slowly to user requests, and the application windows turn white. It appears on the Proxmox interface that the CPU load jumps to 100%.
windows 2019 standard rdp server is very slow
- Thread starter tamas91
- Start date
A couple of questions: how many users are we dealing with here, and what is the link speed of the network adaptor?
I take it you installed the virtio drivers for windows (like from virtio-win-0.1.229.iso)
I'm having difficulty finding it, but is that an HDD RAID array, or an SSD one? [physical host]
I take it you installed the virtio drivers for windows (like from virtio-win-0.1.229.iso)
I'm having difficulty finding it, but is that an HDD RAID array, or an SSD one? [physical host]
Dear Nuke Bloodaxe! The physical host is a Dell R620 server, with 256 GB RAM, an SSD RAID 10 array, and Samsung Pro SSDs inside. I've installed the virtio drivers, and 30 users are using the machine. Its physical Ethernet speed is Intel Gigabit Ethernet.
Attachments
Samsung Pro = consumer SSDs. Not recommended for production use. Is this a hardware or ZFS RAID? As I could see from your screenshots you’re running the VM on your local-lvm and SSD emulation is unchecked.
Bottlenecks should not be CPU and RAM, but your consumer SSDs. When plenty of users work on the server, the SSDs will run out of cache. Just my guess.
Bottlenecks should not be CPU and RAM, but your consumer SSDs. When plenty of users work on the server, the SSDs will run out of cache. Just my guess.
there is a hardware raid on the machine with a dell raid controller, Windows 2012 was running before that and there were no complaints
hw raid disable disk cache to allow hotplug and consistency.
hw raid has its own cache (=bbu) protected by its own battery to allow regular/accelerated performance.
pve, by default, iirc, bypass the cache, so you need set Cache to "Writeback cache"
btw, I'll not confident with Samsung Pro / Consumer SSDs behind a hw controller because, iirc, trim isn't supported. that's why SSDs for server aren't cheap, they're over provisioned + PLP help to optimize writes.
hw raid has its own cache (=bbu) protected by its own battery to allow regular/accelerated performance.
pve, by default, iirc, bypass the cache, so you need set Cache to "Writeback cache"
btw, I'll not confident with Samsung Pro / Consumer SSDs behind a hw controller because, iirc, trim isn't supported. that's why SSDs for server aren't cheap, they're over provisioned + PLP help to optimize writes.
I noticed that when users complain that rdp freezes, the ping response time increases to 400-500 ms, I reinstalled the drivers but it did not solve the problem