RTO Software have released a whitepaper in which they provide measurements of the impact of installing Citrix Presentation Server 4.0 on the performance and scalability of Windows Server 2003 Terminal Services. It turns out that the impact on both scalability and performance is quite severe. Wilco van Bragt had this to say about it:
I did not expect the results of their tests, so definite worth reading.
If these results surprised an SBC expert such as Wilco, I’m sure they’ll surprise a great number of people. I myself was not so surprised – this performance and scalability hit is a direct result of the Citrix Presentation Server architecture.
With the Citrix solution you install everything, the entire package, on each and every Terminal Server in the farm. Moreover, each Terminal Server does its own license management (through direct communication with the license server). Each Terminal Server communicates with the Zone Data Collector and on occasions, with every other server in the zone. Sometimes it even communicates with other servers outside the zone. It also connects to the IMA Data Store (directly or indirectly) and provides an XML service for the Web Interface. Is it any wonder, then, that so many resources that could have been allocated to Terminal Server session are consumed by the Citrix infrastructure itself?
We designed PowerTerm WebConnect very differently, separating all the management infrastructure into an independent component, a connection broker, that is wholly distinct from the Terminal Servers. Onto the Terminal Servers themselves all that is installed are 1.5MB worth of files. No changes are made to the Terminal Servers logon process, and no standard Windows components are replaced. As a result, PowerTerm WebConnect does not degrade Terminal Servers performance or capacity.
For more information regarding this architectural difference between PowerTerm WebConnect and Citrix Presentation Server see my BriForum presentation “Two-Tier Versus Three-Tier SBC Architectures”.