Though View 5 has yet to launch, VMware said this week it will improve its PCoIP remote desktop protocol for virtual desktops by reducing bandwidth consumption on LAN and WAN connections. While this fixes one problem, IT pros say PCoIP performance issues remain.
IT pros using VMware View with PCoIP say it is a bandwidth hog that performs inconsistently, particularly over a WAN. While VMware's newly disclosed improvements to PCoIP may address the WAN/bandwidth issue, and it may not satisfy VMware's power users.
The latest PCoIP improvements from VMware and Teradici Corp. increase network user density and reduce overall bandwidth usage on both LAN and WAN connections, according to VMware's blog. VMware said that internal testing of these new capabilities results in up to 75% bandwidth savings. Of course, actual savings depend on the environment.
The new PCoIP performance settings include a configurable client-side cache control that admins can use to enable, disable and configure the amount of cache used. VMware View will cache images and portions of the desktop to minimize retransmission of pixels across the network to reduce session bandwidth. This feature will mainly be for static screen content that doesn't need updating, or in situations where only a portion of the virtual desktop needs to be refreshed.
Another improvement is the ability to disable "build to lossless" in PCoIP. Build to lossless lets IT send lower-quality snapshots faster and lets end users access their View desktops from remote devices easier. Now, IT will have the option of disabling build to lossless in favor of build to perceptual lossless, which improves bandwidth consumption.
PCoIP in View 5 also includes an update to the lossless CODEC used for certain types of text. This update improves compression and results in lower bandwidth requirements.
What about PCoIP performance?
One of Citrix Systems, Inc.'s advantages with its XenDesktop/HDX protocol has been lower bandwidth consumption and good performance over WAN compared to View/PCoIP. VMware said in a blog post that the latest PCoIP bandwidth improvements give it parity with Citrix.
But the bandwidth improvements don't address underlying PCoIP performance issues that stem from the way the protocol works, said Ken Fanta, IT director for the Department of Orthopedics and Rehabilitation at the University of Wisconsin.
Fanta uses VMware View and PCoIP over a LAN using 10GbE and Fibre Channel and host servers that don't have a high CPU load, he said. Still, PCoIP cannot provide the performance levels his end users need to run graphics, videos, flash drives and X-ray-related software.
"We tried turning off lossless and degrading everything that could be degraded, but it all came down to the way PCoIP works with the CPU," he said. "The data just can't get processed fast enough for power users doing graphics."
PCoIP performs all of its encoding on the remote server host and transmits a series of bitmaps to client devices. This means all of the processing is performed by the host CPU, which increases CPU loads.
By comparison, Microsoft's RDP enhancement -- RemoteFX -- works the same way, but handles the data differently. Microsoft requires the remote host to use a graphics processing unit (GPU) to perform bitmap encoding so CPU resources aren't overloaded. Of course, the GPU add-on card can be expensive, so the better method depends on the individual situation.
To get greater VM density per host, there has to be less resource consumption on the host side. IT pros can alleviate CPU load issues by purchasing Teradici's APEX 2800 PCoIP server offload card, which provides better density and more consistent performance, regardless of the overall demand on server CPUs.
One IT pro said VMware had two big things to fix with PCoIP, and the company chose to fix the one that doesn't alleviate host resource costs. But, he said, the bandwidth improvements are significant.