RDP vs RemoteFX

The group of enterprises “X” uses terminal servers.
The new season began and in one of the representative offices cpu load began to reach 100 percent, which is bad, especially after users began to complain about the speed of work.
The cause of the problem was not clear, the number of employees did not change, the software did not change ... All offices in the same conditions.

I assembled a test bench and began to look for a solution ...
For a long time I went through different settings of the server and client places, this is a separate topic.
In a creative search, I compared the RDP and RemoteFX protocols, I decided to publish the results.

HP ML350 G6, 1 * Xeon5620, 42gb RAM.
DirectX hardware graphics card is missing.
HP MSA P2000 G3 SAS, from 4 SAS disks the R5 array is assembled.
ESXi 5.1 is installed on the server.
Terminal servers are VMs, 4 vcpu (8000 MHz) and 20gb RAM are allocated, Windows Server 2008R2 SP1 is used as a guest OS.

The processor load was compared in three applications: IE11, Adobe PDF Reader 11, 1c8.
I did 8-10 measurements, at the time of measurements only the experimental user and the administrator user worked on the server.
I used two laptops with Windows XP and Windows 7 SP1 as client places, and an HP t510 thin client with installed HP Smart Zero 4.4 OS.


IE11, a test video was launched, which was located on youtube.
RDP - CPU load 21-23%
RemoteFX - CPU load 11-18%

After replacing laptops with an HP thin client.
RDP - CPU load 17-21%
RemoteFX - CPU load 10-12%

In laboratory conditions, the difference was 5-10% CPU time in favor of RemoteFX.
I will add that RDP is not even close to RemoteFX by video playback smoothness, with RemoteFX turned on, at first glance, there is no difference compared to a regular PC.

He decided to conduct all further measurements on the thin client HP.

We pass to the PDF document and scrolling.
RDP - CPU load 16-20%
RemoteFX - CPU load 12-17%
The difference in favor of RemoteFX was 3-4%.

It’s 1s8’s turn, again we will scroll through the list of documents.
RDP - CPU load 14-17%
RemoteFX - CPU load 17-18%
The difference in favor of RDP was 1-3%.

Honestly, I did not like the result of 1c8. I decided to check everything and make additional measurements.
I re-measured the results, everything seems to be ok, I fit into the error in the measurements, about 1-2%.
The results of 1s can be attributed to a measurement error, in the end it turns out that 1s does not matter how the user connects - by RDP or RemoteFX.

If you sum up the preliminary results

I think that video is best suited for assessing the quality of the codec, I decided to conduct other tests, since working with 1s and documents should take up most of the users working time.

I used to try looking at PCoIP, I didn’t like the result, maybe I need to look again, but don’t worry, RemoteFX will cost less than PCoIP, and I like the VDI concept less than terminal servers.

In case of enterprise “X”, 18-24 users work on one Xeon5620 processor with a load of 40-80%, and in addition to the terminal server, the domain controller and some other small vm work.

As I see it, the implementation of RemoteFX will reduce the load on the server processor by 20-30%, or will add about 5-7 users.

Interest in RemoteFX began to grow, and measurements decided to continue

First, we will compare how the increase in the quality of the transmitted sound affects.
In standard settings, sound quality is selected dynamically, when a sufficient number of users are working on the server - this is audible.

We watch the video on YouTube (RDP), the processor load is 18-22%, with standard settings the result is 17-21%.
We watch the video on YouTube (RemoteFX), the processor load is 10-16%, with standard settings the result is 10-12%.

I conclude that the difference is minimal and, if desired, you can safely set high quality.

However, I ask you to pay attention to network traffic, I do not measure it, all users and servers are located at a distance from the switch; if you work on a narrow channel, you will have to consider network traffic.

Next, how RemoteFX will work when changing settings, frame rate, picture quality, codec optimization

Screen capture rate = Lowest
Screen Image Quality = Medium (default)

YouTube clip:
CPU load 5-8%

PDF and scrolling:
CPU load 14-18%

1s8 and scrolling:
CPU load 12-18%.

In the case of media, we get a win, but it is immediately noticeable that the video does not play so smoothly and jerks are visible, similar to those with RDP.
If you think about it, there is nothing wrong with that, it all depends on the tasks that users must perform.

Although in the case of office work, meaning is lost, working with documents consumes twice as much processor time.

Screen capture rate = Medium (default)
Screen Image Quality = Lowest

YouTube clip
CPU load 5-11%

PDF and scrolling
CPU load 12-16%

1s8 and scrolling
CPU load 18-21%.

We get the result where there is no gain for office tasks, and for media, the result of network traffic may be visible.

Screen capture rate = Lowest
Screen Image Quality = Lowest

YouTube clip
CPU load 5-10%

PDF and scrolling
CPU load 12-16%

1s8 and scrolling
CPU load 16-19%.

No comments.

Screen capture rate = Highest (best quality)
Screen Image Quality = Highest (best quality)

YouTube clip:
CPU load 9-13%.

For the results for 1s and PDF, as it turned out, I did not have the patience.

From the Highest (best quality) settings, I expected a different result, and the result can be attributed to a measurement error.

Next up is the codec optimization, Text vs Rich Multimedia

The default setting for the Rich Multimedia codec.

PDF and scrolling:
CPU load 13-16%

1s8 and scrolling:
CPU load 16-19%

Summary table


On all terminal servers I turned on RemoteFX, it definitely won’t be worse.
It became interesting how the result will change if you add a hardware graphics card.

Also popular now: