Tom Valovic posted an interesting article about an interview with Stuart Robinson of Teradici on the purported benefits of Teradici’s PC over IP protocol licensed by VMware.  Tom reports on a Teradici claim that PC over IP has a lead over ICA in that it has an ability to dynamically adjust the bandwidth needed by the remoting protocol, in response to available network bandwidth.  Specifically, Robinson (who actually knows better) claims that ICA has no such ability. 

Wrong.  Citrix ICA has had an ability to dynamically adapt its coding algorithm not only in response to availability of network bandwidth, but also in response to available encode/decode capabilities at the server and client side respectively, for over two years.  This allows ICA to deal with complex rendering problems with grace and to deliver high fidelity across a highly disparate set of server/network/client combinations, dynamically adapting as system conditions change. 

Of course PCoIP is interesting, but it’s “just another protocol” with small user base.  It’s not even particularly useful as a software based encoding algorithm, which seems to indicate that VMware is rather desperate.  Moreover, conversations I’ve had with the OEMs who adopted Teradici chips to hard code PCoIP into the server, that approach is proving to be a difficult sell to customers, since it  ties the server forever to a particular (version of a) delivery protocol from a particular vendor in a narrow proprietary architecture, and moreover it is really only useful in tethered enterprise LAN based configurations today.   In contrast, in both XenApp and XenDesktop ICA is an optimized software only solution and runs superbly on any server, meaning that the server can be repurposed at any time.  Moreover, when running Microsoft TS / XenApp virtualized on XenServer, the architecture has been independently shown to be 70% more efficient than any other virtualization platform.    Just the benefits in terms of number of users per server with XenApp on XenServer would completely obviate the need for any hardware based encoding.    Bottom line: if your hypervisor can’t offer the raw performance needed for remote delivery protocols such as TS/ICA, licensing a hardware-optimized protocol in the hope that it will work well in software definitely is not going to change the game much.