Good afternoon everyone.
I have a client / server application running with Progress 11 (11.6.3).
Users connect in terminal server (balance with 4 servers).
TS Servers run on Windows 2012R2.
Each server has 32GB of memory and 16VCPU.
On average 15 users live on each server.
When running a program on these servers this program takes 30 min to execute.
This same program on another server (but running with 2008R2 on a server with 6GB of memory and 2VCPU) takes less than 5 min.
All servers are virtual.
Is there a problem reported with win2012?
Is the VMWare or HyperV?
If VMWare, what is the NIC type? E1000 or vmxnet3?
Paul
Also -- 16 VCPUs? Do you *really* need 16 vcpus? If you do... does the physical host have 16 physical cores free when you need them? Or is your hypervisor spending a lot of time waiting for those rare moments when there are 16 free cpus to run your VM on?
If you look at perfmon and see that you're only using 10% of those 16 vcpus try scaling back the VM to 2 cores (or whatever "right-size" is a good close match to actual cpu needs). You just might be amazed at how much more you can get done by demanding less.
You might want to look at disabling "Fair Usage" on your servers. This is a new feature and it is enabled by default. I know from experience that this can cause performance problems. With the amount of VCPU's you have and the amount of users on each server, I can't see any need for you to have it enabled. www.synergex.com/.../
|
|
About 5 years ago [mention:44a028c96ca44788b729e5185220e84a:e9ed411860ed4f2ba0265705b8793d05] saved my butt with this vmxnet3 vs E1000 tip. I never forgot it.
|
I don't know the solution for Hyper-V but if you find, please share!
Regards,
Paul
Hi Paul,
There is a know issue with the Broadcom netXtreme network adapter that caused bad latency in Hyper-v. Here is the link: support.microsoft.com/.../virtual-machines-lose-network-connectivity-when-you-use-broadcom-netxt