October 18, 2011 | Written by: Andreas Groth
Share this post:
This Real-World VDI series has five parts: (Part 1) Multi-vendor stacks, (Part 2) Advanced graphics, (Part 3) Cost case challenge, (Part 4) Recent announcements, (Part 5) Future of VDI
Parts 1 – 5 are also posted at Virtualization Matrix.
The announcements at VMworld caused me to adjust my view of VMware’s user capabilities and vision.
So what did VMware announce and why will it impact the current pattern and future directions?
There are two streams:
- The immediate impact with the latest VMware View 5/vSphere 5 product announcement
- The longer term impact with the announcements around Project Horizon, Project Octopus, ThinApp factory integration and the (potentially amazing) AppBlast
The first stream is relatively straight forward but has immediate impact (as available now).
So what has been announced with View 5?
- PCoIP enhancements reducing bandwidth (up to 75%) and increasing latency tolerance (up to 200 ms)
- Support for Direct x 9 and OpenGl 2.1 using a virtual 3D graphics adapter
- View Persona Management, which integrates user environment capability (as discussed in Part 1, Multi-vendor stacks) to manage user personalization when deploying stateless desktops
The promises on the PCoIP improvements will have to be proven in real-life environments but it is clear that VMware is focused on addressing one of the major customer concerns aggressively. The profile management was long overdue (and therefore welcome) and will utilize just-in-time retrieval of profiles to minimize the transfer impact. It will not address physical systems or allow profile management across different operating systems, so third-party apps such as AppSense and others will continue to provide value (but might not be required).
So what is this 3D function? vSphere 5 allows you to create a virtual 3D adapter in the vm (virtual GPU), which essentially emulates the behavior of a physical GPU. This function has the great advantage that no physical GPU is needed and that you can run multiples of those 3D capable VMs on a single host (achieving potentially higher user density than HDX-3D, which has a 1:1 mapping between user and GPU, at the same time supporting a wider range of graphics workloads than RemoteFX – again, all without the requirement for a physical GPU).
VMware’s virtual GPU has the advantage that multiple users can share a single host for 3D delivery and no physical GPU is required in the server.
However, because the host has no physical GPU to do the work, the CPU ends up executing, which can create significant overhead. How much will depend on the nature of the graphics workload, but in offline discussions we have heard that “it can potentially half the number of users per host.” We’ll have to wait for sizing guidelines (or test ourselves).
A secret weapon here might be the PCoIP offload adapter from Teradici, which allows offsetting that overhead significantly. I’ll cover this in another post in more detail because we have already done testing internally, but the card will not GA for a while. Alternatively, if you go to VMworld Europe, come round to the IBM booth and I’ll show you that setup live!
Should we expect VMware’s PCoIP/virtual GPU to deliver better high-end graphics capability than HDX3D?
I see it as a more flexible solution for “casual to medium” graphical workloads but expect, for example HDX GPU Passthrough, to still have the edge when it comes to high-end graphics (with the trade-off in user density).
I would certainly expect VMware to be keen to deliver (physical) GPU offload and client-side offload in a future release to complement the current capabilities.
VMware has certainly cleared some major road blocks with the latest release and we should expect increased interest in View-based deployments
Note that the 3D feature has a dependency on vSphere 5. That means that you can’t get this functionality with earlier vSphere versions (although View 5 will support them). More importantly, VMware cleverly introduced a “vendor stack affinity,” which means for example, you will not get the 3D functionality with XenDesktop on top of vSphere (if this has been your ‘stack of choice’).
Going forward, I expect to see more of this “vendor dependency” approach that will drive arguments for a “single-vendor” stack. Another example is Citrix’s IntelliCache, which depends on XenServer and VMware’s View Accelerator or “CBRC,” which depends on vSphere 5 (the integration of CBRC think “IntelliCache for vSphere” – has been delayed and is now a “future enhancement” for View).
If we review the graphics capability table now, we see that the picture has changed significantly.
Overall, VMware has clearly addressed some of its weakest points in an impressive way with the latest release.
Does that mean VMware will displace Citrix and take the crown? Well, only time will tell but there are more aspects to this then “just” the product updates to View and vSphere:
- Nothing will happen overnight, VMware will need to provide proof points for actual performance, scaling and overhead. The code is new so many will wait and see (regardless of VMware’s track record for code stability).
- Citrix user heritage and customer affinity is strong.
- Citrix’s overall user portfolio (over and above VDI) is still considered unmatched by many as of today (XenApp, branch repeater, NetCcaler, Citrix Receiver portability, and so on).
Saying all that, VMware has certainly cleared out several major road blocks with this release and I’d be very surprised if we don’t see an increased interest in VMware View-based deployments.
I for one certainly consider View 5 a far more attractive proposition than before … and see a good potential for HVD-based deployments.
As mentioned, beware of increasing stack dependency for advanced functionality when using multi-vendor stacks and evaluate carefully.
So what about the second “stream” … how is the vision that VMware (and others) create for end-user computing impacting the future of VDI? Have we really reached the “end of the PC era?”
Read more if you are interested: “Real-World VDI (Part 5): End-user computing, the vision – Horizon, AppBlast, Octopus, and “Should I skip the VDI bubble?””