Lately, I have been getting a lot of queries around sizing RDS servers for Horizon Applications.
Though there is guidance around RDS sizing in the VMware Horizon 6 Reference Architecture Technical White Paper, there still seems to be lot of people requiring a simple tool do so.
Page 22, 23 and Page 25 of the white paper covers RDS sizing and provides you with details.
The article provides a sizing spreadsheet which by changing 5 cells you can determine the number of ESX hosts required for your Horizon Apps RDS servers.
The aim of the sizing spreadsheet is to make things simple.
The only way in my opinion to accurately size RDS servers is to actually test it as there are so many factors that affects user density and scalability.
Some of them are:
- Clock Speed of the core. 2.4 Gz core offers better performance than 2.0 Gz core
- Number of sockets. 24 cores on a Quad core ESX host usually is less performant than 24 cores on a Dual core ESX host.
- Applications types. Outlook 2016 with O365 with large OST files takes up a lot of CPU, memory and IOPs.
- Application usage,
- OS version. Windows 2008 R2 offers better user density than Windows Server 2012 R2 which offers better user density than Windows Server 2016
In the absence of real testing, the next best option is to do an application and workload assessment with tools like Systrack.
So assuming that we don’t have Systrack data and do not have the luxury of testing the load, my sizing spreadsheet should provide a conservative output for your RDS sizing.
Here it is: https://www.dropbox.com/s/jc1zryijswl03tl/RDSH%20Sizing%20v1.0.xlsx?dl=0
Here is how a 2000 users RDS sizing looks like:
- Blue Boxes requires your input
- Purple Box is the output which is the Total number of ESX hosts.
- No overcommit for vCPU to pCPU. 1:1 mapping.
- Ratio is nothing but a percentage. 20% of heavy users = 0.2, 50% of medium users = 0.5, 30% or light users = 0.3
- VMware recommends 30 users per RDS server. Also recommends 4 vCPU per RDS server.
- Buffer Ratio is up to you: for 10%, the ratio is 1.1. for 30%, the ratio is 1.3 and so on.
For the simplicity, I have excluded IOPS and Storage. If you need a number you may assume 10 IOPS for heavy user, 5 IOPS for medium user and 3 IOPS for light user.
Hope you find this useful.