Teradici / VMware will demo soft-to-soft PC-over-IP at VMworld 2009. Boom or bust?

One of the big desktop announcements from last year's VMworld was that Teradici and VMware signed an agreement to work together to develop a software version of Teradici's PC-over-IP remote display protocol for inclusion in VMware's desktop virtualization products.

One of the big desktop announcements from last year’s VMworld was that Teradici and VMware signed an agreement to work together to develop a software version of Teradici’s PC-over-IP remote display protocol for inclusion in VMware’s desktop virtualization products. After a year of work with limited public demonstrations (video), the two companies will publicly demo their “software-to-software” implementation of PC-over-IP at VMworld 2009 in a few weeks.

I’ve written about PC-over-IP quite a bit over the past few years. For those who aren’t aware, PC-over-IP—or simply “PCoIP” for those coolios in the know—is a relatively new remote display protocol that offers a perfect user experience. The only catch is that it requires custom hardware on both ends (typically in the form of a PCI plug-in card on your remote host and a chip that’s built-in to your thin client device on the client end). This wouldn’t be too bad in-and-of itself, but unfortunately today’s PCoIP chips are single session only. So if you want to use them for VDI, you’d need an external chassis with about 50 PCI slots in it. :) The other traditional downside to PCoIP was that it didn’t do so well over the WAN, although the 2.1 firmware release addresses that quite well.

So people love PCoIP, except for that pesky hardware requirement.

Enter VMware.

VMware has struggled in the remote display protocol space. They’ve waffled between “RDP is good enough” to “we are protocol agnostic” to “we’re licensing certain components of TCX from Wyse” to “Screw it, we’re developing our own! (via the Teradici announcement). Okay, technically VMware is still agnostic, but we’ll see how long that lasts.

So anyway, last September VMware and Teradici announced that they’d work together to create a software-only version of PCoIP. Not too much is known about the details of this apart from (1) the deal is non-exclusive, so Teradici could do this (or be doing this) with other vendors, and (2) the software and hardware implementations of PCoIP will be inter-compatible in either direction. (Which is great news for the six people using it today!)

That said, folks are skeptical as to how well the software PCoIP will work. “Umm hello?... didn’t Teradici built custom chips for a reason? If this was so easy, wouldn’t ICA or RDP be doing it now?”

So what do you think? Will the software-to-software implementation of PCoIP suck? Or will it rock?

The soft-to-soft PC-over-IP will suck!

PCoIP was conceived, designed, built, architected, and implemented around the concept of having custom ASICs at each side of the connection. Sure, you can emulate that in software, but to what end? You’ll peg the CPU on the host, killing your user density.

The whole concept of a PCoIP implementation in software is dumb, and in a few more years we’ll joke over some beers about “that one time Teradici tried to build their stuff in software.”

The soft-to-soft PC-over-IP will rock!

You think Teradici isn’t aware of the challenges of implementing this thing in software? You think they didn’t consider that before signing the agreement with VMware?

True, ICA and RDP have something the PCoIP doesn’t: A long legacy. RDP and ICA were designed more than a decade ago when peoples’ requirements were much different. Even though PCoIP was initially designed for custom hardware, its designers built it from the ground-up for the full desktop experience: multimedia, multiple displays, USB redirection—all this was designed into the protocol from the beginning instead of being hacked in later.

So they’ll be fine, just you watch!

Place your bets

What do you think? Will the software version of PCoIP suck or rock? (By the way, remember the “user experience, low bandwidth, and CPU load: choose any two” ruleset when thinking about remote display protocols.)

Your thoughts?

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

Even if they fix it... what will be the next bottleneckt then? The VDI server's GPU?

I reckon you'd need a lot more of them. So even if they 'solve' it they'll just bump into the next bottleneck.


Charlie comments are totally on point. At the end of the day SW takes up CPU resources and memory resources. Unless there is some sort of blended model I can’t see this succeeding. Take a look at companies out there like HP who have RGS. That’s software to software and it has its weaknesses just like other vendors. My real question is when are we going to see VMware partnering with Nvidia for a board full of GPU’s that could be assigned to a vm or group of VM’s.


What I have been saying for a LONG time resumes all this: unless a FUNDAMENTAL change happens on the hardware side (like multiple GPUs exposed to the hypervisor, etc) you will always have to find a way to do in software what should be provided by the hardware layer. This kills density as pointed above.

I still think the best thing that could happen for VDI or Server Centric technologies for that matter would be Intel buying Citrix... Just read my post (http://wtslabs.com/blog) on that topic. :-)


2 points on Adam's comment.

a) GPU's are NOT required for VMs today for mainstream enterprise users.  Keep in mind that 2D "business graphics" are natively rendered in the CPU for Vista.  

GPUs in VMs will be required in the future to support Aero and heavy 3D user requirements. Selling GPU's for VM's to support Aero is a great growth opportunity for NVIDIA, AMD etc as they don't sell GPU's for these users today.  

b) I agree a blend is necessary - IT can consolidate any user with a blend of PCoIP hardware (1:1) and software (N:1) technology in the same way IT have a mix of low-end, mid-range and workstation desktops today.  Now this changes to a highly virtualized desktop (large N), moderately virtualized (medium N) or 1:1 blade/rack PC.

As noted above, VMware View with PCoIP software technology (no hardware required) addresses mainstream enterprise users. Users with Aero/3D requirements can be consolidated using PCoIP hardware that does not load the host CPU at all.  PCoIP hardware products are available from Dell, IBM, Fujitsu, Samsung, Amulet-Hotkey, Boxx, Clearcube, DevonIT, ELSA (Japan), EVGA (America's & EU), Leadtek (Asia), Verari and more to be announced shortly. This allows consolidation of all users - which is a key requirement we hear from IT all the time.  

While IT do not require Aero support to consolidate their enterprise users today, it will be a requirement in the future.  I agree GPU support in VMs will be necessary for Aero & other 3D users – I am keenly looking forward to support of GPU's in a virtual desktop environment - the better the image rendering allows me to show the full power of PCoIP technology. From a technology point of view, PCoIP protocol can support 3D graphics once GPUs can fit in a VM environment – its just a display’s worth of pixels for us to compress whether a basic 2D display, Aero display or full 3D graphics – still just pixels that PCoIP technology can handle.  


I like the fact that you are pointing out the CPU load. When discussing VDI we must always be cognizant that all of our drivers are "virtualized" - i.e. software. This becomes a very big deal when we are pushing sound or graphics through the channels. In the case of PCoIP I suspect there will be a significant impact on both guest and client CPU utilization.

Higher CPU utilization could lead to less user density per server, this is assuming many users are doing a graphical or sound oriented task at the same time. The newer processors will certainly help out.

My opinion is that PCoIP will rock on the LAN when mated to the right server hardware. I don't think repurposing your old DL585 is necessarily a good idea here. I also think that PCoIP will continue to have it’s issues with Latency/Jitter/Packet Loss (Just guessing here, but eventually we might see have similar technologies like that in WAN acceleration devices built into to the VM driver packs and clients. Much like what citrix is doing with “Turbo Charge” but on a much more software based scale.) I think in years to come, software PCoIP will not be like ICA/RDP, I will either die or morph back into hardware, like the others have stated.

I also think we will eventually see this come back to GPU hardware that is optimized for a Hypervisor. Think of a gaming machine with Hypervisor specific enhancements just for VDI. Hmm, a Quad Eight Core CPU, 1 TB of RAM with 8  to 10  GPUs and Sound processors right on the board with an ESXi like Hypervisor with direct access to those hardware components all bundled in a nice 2U appliance! (OK there is my day dream for the day..)


@Shanetech - re: PCoIP will continue to have it's issues with Latency/Jitter/Packet loss ...

You are correct - ALL remoting technologies will always have issues with these limitations and constraints because they are like forces of nature you cannot control. The only thing you can do is react to them.

Based on extensive testing of many of the different remoting protocols, I can go out on a limb and state that PCoIP has 'THE BEST' reaction and management of those forces of nature, and they are continually improving (WAN is now supported). ICA and RDP interact with the networking protocols at a higher networking layer and that has its issues and complications for providing the best experience. They ‘state’ they react to the network, but looking at how you configure them, they react ‘at best’ with couple of networking bandwidth thresholds which causes different types of compression to be used (ICA SpeedScreen), and the norm is you need to set up what your ‘forces of nature’ will look like before you even connect (RDP – Performance settings ->Connection speed). In addition, once a session is established, other protocols ignore the way the networking topology changes as more users are sharing a single networking bottleneck. PCoIP was developed from the ground up by developers that have a massive amount of 'networking and ethernet' experience (Just look at the Mgmt team). The PCoIP protocol melds the networking layers directly into how it decides what to use for compression with a much wider and more granular scale compared to other protocols. It also reacts in real-time to the three forces of nature – assuming that the network topology and usage is dynamic. This is why there is a myth that “PCoIP” is a networking bandwidth hog. It consumes as much bandwidth as it can ONLY when it can. If other traffic is operating on the same network, it backs off immediately to allow that traffic to pass through and re-adjusts itself to ensure the best End User Experience, while continually attempting to grab back as much bandwidth as possible.


I have no doubt that PCoIP will be successful for customers that have specific needs.  I do find it very humorous though that for the last 2 years the Teradici guys have jumped on every single display remoting post to rip on software display remoting and mention how superior their hardware approach is.  Now that there's a software version of PCoIP, it's all of the sudden become cool. :p



Ohh and OUCH! on this comment:

"(Which is great news for the six people using it today!)"



@Shawn Bass - I appreciate your point of view; however, I did not mention in this post anything about how the hardware approach was superior to software display remoting.  I was strictly speaking about PCoIP remoting technology (hard or soft).

But I will do it now :).

I still believe that PCoIP hardware remoting solutions are superior to all software (including PCoIP) remoting solutions, when you are measuring User Experience.

Software PCoIP solutions opens up the use cases for the PCoIP remoting technology to users who require access via different endpoints (clients), such as Laptops, refurbished desktops, or even some form of thin clients.  There are those users where 'quality of remote experience' is #2 on the short list of requirements.

Oh and I also had to laughed at the jab Brian gave on the number of deployments  (@Brian - you've been on a tear lately).  On that - It would be interesting to see the growth of any/all of the remoting technologies in terms of full desktop replacement, and how they stack up to each other chronologically as well as market share of 'desktop replacements'.



@Paul - I wasn't referring to comments made in this article.  I was referring to general comments made by Teradici folks over the last few years where generically speaking "Software protocols are crap and hardware is where it's at".  I'm not saying that I don't get it that you see a need for both.  I'm simply bringing up that you guys were raining on everyone's parade about software rendering over the last 2 years and now it's pretty cool ;)



@Paul Helter, Why the hell do the dummies at VMWare think they can win a Desktop battle against Microsoft when they own the OS? You can't win that with an immature software protocol that eats bandwidth and when MS hates your guts. What will you do when Calista is not supported on ESX? Will VMWare all of a sudden offer support for Hyper-V. VMWare also just bought a Java company for 420M, so they hate .NET apps too. VMWare can't win the Destkop, it's not what they do, they are becoming a systems management company as much of press sees. They are now competing with MS, Citrix and the systems mgmt vendors, Oracle hates them too. They have no friends, so how the hell are they going to win in all these areas? I certainly don't buy they invest anywhere close to their core business, how does that compare to their competitors? PC-IP will is just eye candy that is designed to further confuse people.


I think Paul Helter hit on a key factor in determining what kind of graphics performance you can expect - compression. I've configured PCoIP directly between host workstations and remote clients, and also through network swtiches / routers. There's a world of difference in performance between these configurations. However, my laptop already has the rdp client built into the OS. That's a solution that I don't think Teradici will be able to top anytime soon.

In general, I think a dedicated PCoIP hardware solution tops a software-based solution; much like a hardware-based MPEG decoder usually beats a software-based MPEG decoder solution. As always, the tool used depends upon the application and environment it will be used in.


It's just a shame that all they demoed at the keynote was a slide show with some fade effects.  Not too exciting if you ask me.  And what about desktop virtualization in general?  It sure seemed like they are putty a lot of focus on the cloud and very little on the desktops.


Despite the lipservice VMware gives, VMware is not that committed to Desktop Virtualization.  It's evident in the weak demo on PC-over-IP delivered at the keynote. It's evident in the way View is being sold by their field and channel partners. It's evident in the number of really talented VMware employees who have left the Desktop BU in the last year - like rats on a sinking ship.

Their type I client hypervisor will take a few years to mature after it ships, at least for Enterprise customers. VMware really needs to think about what is the best virtualization solution for customers on the desktop VS. using it primarily as a workload to sell more ESX, VI3, vSphere. PC-over-IP is just another example of how VMware does not truly appreciate what is required to help customers with managing, provisioning, and driving down the cost of desktop computing for business customers.

Just my two cents...