Project Virtual Reality Check have released their Phase 2 white paper on Terminal Server/RDS workloads running on the latest generation Intel processor: the Xeon 5500 series (Nehalem). Besides providing some great figures to support the adoption of Intel’s Nehalem to drive high demand virtualized workloads, this is an interesting and important comparison document for those considering centralised desktop virtualisation.
In their paper, the chaps at Project VRC tested Terminal Server 2003 (x86 an x64) workloads on the latest hypervisors from Citrix, Microsoft and VMware with the latest version of the Login VSI benchmark.
Granted, the white paper focuses on a specific Presentation Virtualization (PV) solution: it primarily considers a Microsoft Terminal Services/RDS solution. Solutions that enhance Microsoft’s core PV offering, such those outlined in our Presentation Virtualization Solutions white paper will add an overhead and so figures will be slightly different. That said, the document does have a short section on the impact of Citrix XenApp’s CPU Fairshare functionality – to demonstrate that CPU throttling and scheduling solutions can have a positive impact on performance for and improve response times for users.
It is useful to have a benchmark for comparing desktop centralisation solutions. Citrix, in demonstrating their New Heights for XenServer Single Server Scalability with XenDesktop adopted Login VSI 2.0 from Log•in Consultants as the core of their testing process. For their comparison, Project VRC used the free Login Virtual Session Indexer (Login VSI 2.1) methodology. I’d suggest that server hosted desktop virtualization vendors should engage this stance, and adopt and help adapt the freely available Login VSI benchmark. Ultimately, it allows a level playing field for customers to be able to compare each solution.
Tests were carried out using a typical server configuration. The document discusses the impact of Office 2007 SP2, Internet Explorer IE7 versus IE8, CPU Fairshare, the impact of the vMMU settings in vSphere, and much more. It is an extensive and interesting document. The test server was an HPDL380G6, 2 x Intel Quad core firstname.lastname@example.orgGHz with 64GB memory and 8 x 146Gb. From the figures presented, a bare metal, x64 2008 Remote Desktop Service server supported 149 concurrent users. Using that server as a host – with HyperV2 or XenServer5.5 – supporting x32 operating system based virtual machines and this increased get @ 160. Their figures state that the VMware hypervisor typically supported @ 15% less users. Interestingly, Citrix recently quoted 125 hosted virtual desktops on a similarly specified device.
The document makes a valid point that running Windows 2003 standard 32-bit Terminal Server workloads without the use of a hypervisor does not make sense any more: Windows 2003 Standard Edition is unable to make full use of the capabilities of the latest hardware. However, the tests only utilise the standard edition of Windows 2003: Enterprise edition is more expensive yet, with R2, supports up to 64GB of memory. Bear in mind, it is possible to support more users without introducing a hypervisor.
That said, Project VRC’s Phase II white paper helps answer a number of important points that you may be asking around how to best deliver Presentation Virtualization services in the future.
- A PV solution appears to support slightly more users per server than hosted desktops – especially if you deploy an x64 PV solution.
- With the new generation of hardware and hypervisors you can now virtualize your RDS/TS servers.
- If you are considering virtualising your existing x32 PV servers – Hyper-V2 and XenServer 5 have better performance on the new generation of servers.
- Most importantly, having standardised testing model gives much better focus for planning and understanding of needs for capital expenditure than relying on vendors’ marketing.