[Paper Introduction] Performance Comparison of PCoIP and RDP in Resource-Intensive Tasks

In my previous post, "Comparison of RDP 8.0 and PCoIP in VMware Horizon 6," I expressed doubts about the validity of the verification methods in the paper I covered. Therefore, I would like to look at another paper that compares the performance of PCoIP and RDP. The paper I'm discussing this time is "Time-based criteria for performance comparison of resource-intensive user tasks in virtual desktops" written by Mikhail Makarov, Prasad Calyam, and A. M. Sukhov in 2014. This paper also references the study I discussed two posts ago.

In this paper, the authors establish the premise that the VDI user experience heavily depends on the communication protocol, and that measuring thin-client protocol performance requires analyzing a complex set of factors, including CPU, memory, and network bandwidth between the server and the client.

In this paper, we create a novel method and a "VDtest" benchmarking toolkit for comparing the performance of common thin-client protocols such as RDP (Remote Desktop Protocol) and PCoIP (PC-over-IP) for resource intensive user tasks involved in graphics-based applications, using a set of "time-based criteria"...

As stated above, they conducted their verification by creating a benchmark tool called "VDtest."

Specifically, they applied an image processing load using an Excel-based application. Rather than simply applying a load on the virtual desktop, they compared the task processing time and resource usage for a series of processes: downloading the target task data to the virtual desktop, the application performing the image processing, and the processing results being rendered and displayed on the client side.

The researchers chose image processing tasks because they believed there was more room for study in this area, whereas previous research had already focused on the load from office applications.

The OS for the virtual desktop was Windows XP. For the communication protocols, they used PCoIP and two types of RDP: RDP from Quest Software and RDP from VMware (making a total of three types: one PCoIP and two RDPs).

In this verification, a USB flash drive connected to the client terminal was made visible to the virtual desktop side via USB redirection. Uncompressed image files from that flash drive were then read onto the virtual desktop, processed for JPEG conversion/compression within Excel, and the images were displayed on the client screen.

Here is the analysis of the verification results using VDtest.

PCoIP in each case exhibits the least run-times than either RDP VMware and RDP Quest...

Looking at it on a run-time basis, PCoIP outperformed RDP.

we conclude that the main dominant criterion component is clearly the video output time t2

The difference between PCoIP and RDP on a run-time basis occurred in the part where the virtual desktop image is displayed on the client screen, with a difference of several seconds appearing in this section. For other parts, such as uploading image files from the client to the virtual desktop or the image processing itself on the virtual desktop, no significant differences occurred.

On the other hand, looking at resource usage reveals another side. When using PCoIP, the CPU utilization stayed stuck at nearly 100%, whereas when using RDP, both versions remained around 50%. Regarding this, the authors describe it as follows:

in comparison to the family of RDP protocols, the PCoIP shows aggressive yet efficient use of CPU, network bandwidth and RAM resources in the context of resource-intensive user tasks in virtual desktops.

From the authors' perspective, while PCoIP uses more resources, they seem to have judged that it makes efficient use of available resources as evidenced by the shorter execution time.

The above is the content of the paper. To summarize it in my own way:

1. In the performance comparison of virtual desktop communication protocols, factors such as data upload to the virtual desktop and display of the virtual desktop screen to the client are included in addition to tasks on the virtual desktop itself. Specifically, the difference between PCoIP and RDP from a performance standpoint occurred in the display of the virtual desktop screen on the client side.
2. From the perspective of resource utilization, PCoIP was higher than RDP, sometimes pegging the CPU utilization near 100%. However, it can be judged that PCoIP has a characteristic of utilizing resources more effectively as it resulted in shorter required times.

I got the impression that this paper is a legitimate evolution of the study I covered two posts ago, as it incorporates evaluation and analysis items for network transfer processing and resource usage—aspects I felt were insufficiently considered in the paper from my previous post. Since this paper was written in 2014 as mentioned at the beginning, I feel that if the author of the paper from the previous post had researched preceding and similar studies before writing their paper in 2017, they could have improved its overall quality.

I think this is a good paper that covers performance measurement points related to virtual desktops more comprehensively and provides insightful analysis of the measurement results, despite being brief. I would have liked to see a video of VDtest in action.

Note that the verification in this paper was conducted by executing an image processing program and evaluating tasks with large changes in the desktop screen. Considering that the situation differs from when using office applications, that the PC used for verification was Windows XP, and that the RDP protocol was quite old, I don't believe this means PCoIP always has an advantage over RDP.