Xendesktop with Citrix Provisioning Server work load analysis

Storage sizing is critical to deploy a virtual desktop solution. Over sizing makes the solution very expensive. Not enough storage ends up bad user experience. 

I have spent last 2 months setup a 750 VM environment to find out Xendesktop PVS IO characteristic with NetApp performance engineer’s help. We used loginVSI work load tool. 

The work load is separated in 4 nodes of a cluster OnTap system. We used NetApp perfstat to study the workload.

Infrastructure VMs are on a NFS volume hosting XenDesktop VMs, PVS VMs, XenApp VMs, Microsoft SQL servers, AD, DNS, DCHP, LoginVSI VMs.

WriteCache NFS volume hosts PVS write cache. 

User data/profile and PVS vDisk is on a CIFS SMB. 

First, IO can be different size (< 512MB, 4K, 8k,..64k), different type IO – read IO or write IO, sequential or random. So we cannot only mention IO by how many IOPS. We need to study IO in different stages.

The table below is the IOPS in different stage and different volume. They are not the scary 100 IOPS for boot. We measure the boot IO by total IO / the boot time.  

The peak IO is about 35IOPS and the average is about 15 IOPS. 

The diagram below shows you the IOPS during boot, login and steady state. Steady state is loginVSI heavy work load. 

Big IO like 64KB can carry more throughput than 4KB IO. The diagram below shows PVS write cache 41% IO is 4KB but 66% throughput is 64K.

If you use NetApp VDI sizer tool – SPM, you should use 64K as the working set.

The best practice for virtual desktop sizing is to access not guess. You can use Liquidware Lab tool, Lakeside tool to evaluate the existing environment. Then you can use NetApp’s SPM to calculate how much storage you need. 

If you don’t have any data, then use the table in the blog as the guidance.