Old equipment RDS farm

[Badaz52]

Hi everyone,

 

We have just had a brand new hyperconverged nutanix solution installed here at the school running VMWARE. This means that I have been left with some old kit that I want to use to create 2x session hosts.

 

I have 3x PowerEdge R710 servers with 64GB of RAM in each so I was going to use the memory to create 2x with 96GB in each. What is the best way for me to install these session hosts? would you just put windows straight onto them or would you install a hypervisor and do it that way? They are connected to 2x SANS but I've been advised that I don't need to use them at all if I use these drives and put them straight into the poweredge in place of the 1GB SD cards they have been booting from.

 

I'm not planning on VDI just straight RDS.

 

Thanks!

[Norphy]

If you're not going VDI, I wouldn't bother putting an extra virtualisation layer on it, I'd just install the RDS Session Host directly onto the hardware.

 

I'd agree with putting the hard disks directly into the server, rather than using a SAN. Its worth mentioning that RDS hosts are very dependent on the speed of their local storage so put it into RAID 10 if you're using spinning rust. It wouldn't hurt to wangle some SSDs to put in there if you can get the budget.

 

Finally, you might want to look into User Profile Disks to store user profiles on, otherwise you may find that your server gets filled up with profiles very quickly indeed.

[Badaz52]

Thank you, this is exactly what I was looking at doing. I have 2x 12 core servers with 96GB ram in each one. How many connections do you think that could handle bearing in mind it is only office applications, emails , sims that they will use.

 

Thanks!

[Norphy]

The general rule of thumb is supposed to be 1-2gb of RAM per user and something like 10 users per CPU core. So in theory, with that much RAM somewhere between 96 and 192 users. In practice, you’ll have to put it in and closely monitor it, looking at disk queue lengths, CPU and RAM usage and network bandwidth consumption.