So let me get this straight, Tolly Grp want me to believe that a 64-bit guest with 14 GB Memory can run more sessions that a 32-bit guest with 3 GB memory, tell me that can't be true!

I am being hopelessly idealistic in expecting that TG might have considered a like for like comparison and run a 32-bit Enterprise Edition guest with 14 GB Memory along side the 64-bit guest, or better yet attempted to run 3 or 4 32-bit SE guests as a more meaningful, and dare I say it real world, performance comparison? 

I am totally lost for words...

The objective of the Tolly Group report was to present independent findings of a study undertaken by Citrix to validate assumptions that enhancements made to XenServer 4.1 would make it more condusive to virtualizing XenApp. Their independent audit included an analysis of both 32 and 64 bit VMs vs. a single physical XenApp server. The report concluded that single 32 bit XenApp VMs running on XenServer performed with 21.5% overhead vs a single physical 32 bit XenApp server, while a 64 bit XenApp VM performed with only 7.6% overhead vs a similar phyical server. There are of course, dozens, if not hundreds of possible permutations from a configuration standpoint to consider, including running multiple VMs. These were not explored in this particular study but are the subject of numerous follow-up studies being conducted right now within Citrix and with customers and SIs. This study served as a basis for creating benchmarks for these new studies to compare their results against.  It was never suggested that customers should only operate XenApp on single 64 bit VMs; but rather that customers should determine for themselves where they are most comfortable in terms of 32 vs 64 bit and their desire to spread their users across multiple instances of XenApp VMs. This was clearly stated in the white paper mentioned in Bryan's post, "Optimizing XenApp perfomance on XenServer 4.1 Enterprise Edition". What was also stated is the fact that a 32 bit Windows OS is unable to utilize more than 3Gb of RAM in a XenApp environment, no longer a limitation with a 64 bit Windows OS.  Our study applied the additional resources available in the 64 bit OS environment which included RAM, page file, and CPU (XenApp in particular performs better with increases in RAM and page file size) to both the physical server and a VM and pointed out the lowest possible virtualization overhead available when all such resources were made available to the VM.

Is virtualizing 64 bit XenApp realistic for all customers? Not likely, but they now have one more incentive to consider migrating to 64 bit than before. Still, virtualizing 32 bit XenApp has demonstrated to have potentially enough benefits for many to consider virtualizing XenApp without migrating to 64 bit. As much as we would like to have produced a one-size-fits-all roadmap to virtualizing XenApp, the reality is that each customer will need to determine their own path. For many XenApp customers, the traditional driver behind server virtualization, server consolidation, might not be the primary driver for them. It may instead be to realize one or more of the other potential benefits available with virtualization, flexibility, managability or availability. The Tolly Group report has pointed out the once formidable roadblocks to virtualizing XenApp, are now going away and the benefits of server virtualization with XenServer are more available than ever before.

Bernie Hannon

Director of Solutions Testing and Certification

Citrix Virtualization and Management Division