Citrix Load Testing Challenges


Today, load testing of a Citrix MetaFrame XP Presentation Server is typically done by executing a server based script using a Winbatch engine or Visual Basic Scripting engine on a MetaFrame XP server. The Citrix Server Test Kit (CSTK) is primarily used to employ this type of testing solution. The benefit of this strategy is cost, as the CSTK is a free product available on Citrix’s website:

However, this scenario is less than desirable for the following reasons:
  • A server-driven solution is intrusive, due to the additional CPU and memory footprint for each started session. Intrusive tests can never yield truly representative results.
  • A server-driven solution is unrealistic as it provides only the server-to-client side of the exchange. No key press or mouse-click information is ever sent to the server.
  • Currently there is no Unix GUI playback engines, thus this solution cannot be applied to MetaFrame Presentation Server for UNIX. A client-driven solution would have no platform dependencies on the server side.
  • Server-side script playback is rendered unrealistic in a thin-client environment due to all client-server events being activated not by events on the client, but by events on the server.
  • Additionally, to instantiate the number of sessions needed to generate the maximum load a server can handle requires numerous client devices with the traditional Citrix Program Neighborhood ICA client. Another challenge in Citrix load testing is to be able to monitor the performance of the system across all the tiers to help isolate bottlenecks wherever they may reside.
Most existing solutions today rely on deploying server-side agents to generate the load which adversely affects the accuracy of the solution and do not measure true client-side performance. The alternative is running a subset of client-side users; however this approach does not provide scaling to production levels.
Also See:

No comments:

Post a Comment