In my previous EUC Layer post I discussed the importance of putting insights on screens, in this post I want to discuss the EUC Layer of putting something on the screen of the end user.
In short, a display protocol transfers the mouse, keyboard and screen (ever wondered about vSphere MKS error if that popped up) input and output from a (virtual) desktop to the physical client endpoint device and vice versa. Display protocols usually will optimize this transfer with encoding, compressing, deduplicating and performing other magical operations to minimize the amount of data transferred between the client endpoint device and the desktop. Minimize data equals less chance of interference equals better user experience at the client device. Yes, the one the end user is using.
For this blog post I will stick to the display protocols VMware Horizon has under its hood. VMware Horizon supports four ways of using a display protocol: PCoIP via the Horizon Client, Blast Extreme/BEAT via the Horizon Client, RDP via Horizon Client or MS Terminal Client, and any HTML5 compatible browser for HTML Blast connections.
The performance and experience of all the display protocols are influenced by the client endpoint device – everything in between – desktop agent and the road back to the client. : for example virtual desktop Horizon Agent. USB Redirected Mass storage device to your application, good-bye performance. Network filtering and poof black screen. Bad WiFi coverage and good-bye session when moving from office cubicle to meeting room.
Who? What? Skip this one when you are serious about display protocols. The only reason it is around in this list, is for troubleshooting when every other method fails. And yes the Horizon Agent default uses RDP as an installation dependency.
Just Beat it PCoIP. Not the official statement of VMware. VMware ensures it’s customers that Blast Extreme is not a replacement but an additional display protocol. But yeah…..sure…
With Horizon 7.1 VMware introduced BEAT to the Blast Extreme protocol. BEAT stands for Blast Extreme Adaptive Transport— UDP-based adaptive transport as part of the Blast Extreme protocol. BEAT is designed to ensure user experience stays crisp across quality varying network conditions. You know them, those with low bandwidth, high latency and high packet loss, jitter and so on. Great news for mobile and remote workers. And for spaghetti incident local networks……..
Blast uses standardized encoding schemes such as default H.264 for graphical encoding, and Opus as audio codec. If it can’t do H.264 it will fallback to JPG/PNG, so always use H.264 and check the conditions you have that might cause a fallback. JPG/PNG is more a codec for static agraphics or at least not something larger than an animated gif. H.264 the other way around is more a video codec but also very good in encoding static images, will compress them better than JPG/PNG. Plus 90% of the client devices are already equipped with a capability to decode H.264. Blast Extreme is network friendlier by using TCP by default, easier for configuration and performance under congestion and drops. It is effecient in not using up all the client resources, so that for example mobile device batteries are not drained because of the device using a lot of power feeding these resources.
Default protocol Blast Extreme selected.
PC-over-IP or PCoIP is a display protocol developed by Teradici. PcoIP is available in hardware, like Zero Clients, and in software. VMware and Amazon are licensed to use the PCoIP protocol in VMware Horizon and AWS Amazon Workspaces. For VMware Horizon PCoIP is an option with the Horizon Client or PCoIP optimized Zero Clients.
PCoIP is mainly a UDP based protocol, it does use TCP but only in the initial phase (TCP/UDP4172). PcoIP is rendered, multi-codec and can dynamically adapt itself based on available bandwidth. In low bandwidth environments it utilizes a lossy compression technique where a highly compressed image is quickly delivered followed by additional data to refine that image. This process is termed “build to perceptually lossless”. The default protocol behaviour is to use lossless compression when there is minimal network congestion expected. Or explicitly disable as might be required for use cases where image quality is more important than bandwidth for example in medical imaging.
Images rendered on the server are captured as pixels, compressed and encoded and then sent to the client where decryption and decompression happens. Depending on the display, different codecs are used to encode the pixels sent since techniques to compress video images can be different in effectiveness compared to those more effective for text.
Blast Extreme without the Horizon client dependency. Client is a HTML5 compatible browser. HTML access needs to be installed and enabled on the datacenter side.
HTML uses the Blast Extreme display protocol with the JPG/PNG codec. HTML is not feature par with the Horizon Client that’s why I am putting it up as a separate display protocol option. As not all features can be used it not a best fit in must production environments, but it will be very sufficient for enough to use for remote or external use cases.
Depending how the pool is configured in Horizon, the end user has either the option to change the display protocol from the Horizon Client or the protocol is set on the pool with the setting that a user cannot change it’s protocol. The latter is has to be selected when using GPU, but it depends a bit on the work force and use case if you would like to leave all the options available to the user.
Display Protocol Optimizations
Unlike what some might think, display protocol optimization will benefit user experience in all situations. Either from an end user point of view or from IT having some control over what can and will be sent over the network. Network optimizations in the form of QoS for example. PCoIP and Blast Extreme can also be optimized via policy. You can add the policy items to your template, use Smart Policies and User Environment Management (highly recommended) to apply on specific conditions or use GPO’s. IMHO use UEM, and then template or GPO are the order to work from.
For both protocols you can configure the image quality level and frame rate used during periods of network congestion. This works well for static screen content that does not need to be updated or in situations where only a portion of the display needs to be refreshed.
With regard to the amount of bandwidth a session eats up, you can configure the maximum bandwidth, in kilobits per second. Try to correspond these settings to the type of network connection, such an interconnect or a Internet connection, that are available in your environment.For example a higher FPS is fluent motion, but more used network bandwidth. Lower is less fluent but a less network bandwidth cost. Keep in mind that the network bandwidth includes all the imaging, audio, virtual channel, USB, and PCoIP or Blast control traffic.
You can also configure a lower limit for the bandwidth that is always reserved for the session. With this option set an user does not have to wait for bandwidth to become available.
For more information, see the “PCoIP General Settings” and the “VMware Blast Policy Settings” sections in Setting Up Desktop and Application Pools in View on documentation center (https://pubs.vmware.com/horizon-7-view/index.jsp#com.vmware.horizon-view.desktops.doc/GUID-34EA8D54-2E41-4B71-8B7D-F7A03613FB5A.html).
If you are changing these values, do it one setting at a time. Check what the result of your change is and if it fits your end users need. Yes, again use real users. Make a note of the setting and result, and move on to the next. Some values have to be redone to find the sweet spot that works best. Most values will be applied when disconnecting and reconnecting to the session where you are changing the values.
Another optimization can be done by optimizing the virtual desktops so less is transferred or resources can be dedicated to encoding and not for example defragmenting non persistent desktops during work. VMware OS Optimization Tool (OSOT) Fling to the rescue, get it here.
Monitoring of the display protocols is essential. Use vROPS for Horizon to get insights of your display protocol performance. Blast Extreme and PCoIP are included in vROPS. The only downside is that these session details are only available when the session is active. There is no history or trending for session information.
There are other options to help the display protocols on the server-side by offloading some of the graphics rendering and coding to specialized components. Software acceleration uses a lot of vCPU resources and just don’t cut it in playing 1080p full screen video’s. Not even 720p full screen for that matter. Higher clock speed of processor will help graphical applications a lot, but a the cost that those processor types have lower core count. Lower core count and a low overcommitment and physical to virtual ratio will lower the amount of desktops on your desktop hosts. Specialized engineering, medical or map layering software requires graphic capabilities that are not offered by software acceleration. Or require hardware acceleration as a de facto. Here we need offloading to specialized hardware for VDI and/or Published applications and desktops. Nvidia for example.
What will those applications be using? How many frame buffers? Will the engineers be using these application mostly or just for a few moments and are afterwards doing work in office to write their reports. For this Nvidia supports all kinds of GPU profiles. Need more screens and framebuffers, choose a profile for this use case. A board can support multiple profiles if it has multiple GPU cores. But per core there only one type of profile can be used, multiple times if you not out of memory (buffers) yet. How to find the right profile for your work force? Assessment and PoC testing. GPU monitoring can be a little hard as not all monitoring application have the metrics up there.
And don’t forget that some applications need to be set to use hardware acceleration to be used by GPU or applications that don’t support or run worse on hardware acceleration because their main resource request is CPU (Apex maybe).
Engineers only? What about Office Workers?
Windows 10, Office 2016, browsers, and streaming video are used all over the offices. These applications can benefit from graphics acceleration. The number of applications that support and use hardware graphics acceleration has doubled over the past years. That’s why you see that the hardware vendors also changed their focus. NVidias’ M10 is targeted at consolidation while its brother M60 is targetted to performance, however reaching higher consolidation ratio’s then the older K generation. But cost a little bit more.
vGPU and one of the 0B/1B profiles and a vGPU for everyone. The Q’s can be saved for engineering. Set the profiles on the VM’s and for usage on the desktop pools.
And what can possibly go wrong?
Fast Provisioning – vGPU for instant clones
Yeah. Smashing graphics and depJloying those desktops like crazy… me likes! The first iteration of instant clones did not support any GPU hardware acceleration. With the latest Horizon release instant clones can be used for GPU. Awesomesauce.
– Enjoy looking at the stars!
Sources: vmware.com, wikipedia.org, teradici.com, nvidia.com