In viewing environments, it confirms that there's no free lunch. My guess is that the performance numbers are reasonably honest. Meircom does reasonably honest work. Could the soft PCoIP performance numbers have been some better with a vExpert there to tweak them, probably. But I also expect that the difference wouldn't be stunning.
The most significant piece of information that was omitted in the report is what type of video cards were used in the HP DL360's that were used as servers. Citrix HDX relies on NVIDIA CUDA GPU cores to accelerate processing. These wind up being a hidden hardware accelerator. Take these out and the CPU and/or bandwidth results would be very different, I suspect.
PCoIP was designed to be a hardware accelerated solution with great compression and encryption. Adapting this to software looks like it has proven to create some inefficiencies. PCoIP in hardware does an amazing job for remote viewing. And requires significant bandwidth. I hope that server hardware acceleration for vSphere arrives soon.
The no free lunch part comes from the three axes of viewer performance (described by Teradici a year or so ago) - bandwidth, computational power/scaling, and end user experience. You can optimize for one, but you have to pay with the other two. If you want great graphics, there's going to be a big requirement for processing, in either hardware or software.
There are lots of viewing environments out there, HP's RGS, NoMachine's NX, RDP in it's various forms, and even the open source SPICE, now from Red Hat. There are compromises for all of these. The best solution is the compromise that works best for you and your end users.
- Eric
If you found this note/reply useful, please consider awarding points for "Correct" or "Helpful"
If you found this note/reply useful, please consider awarding points for "Correct" or "Helpful"
