For 10 months in the cold north, we are addicted to winter weather forecasts (ok, 10 months is a little extreme. Winter does not usually last that long in Minnesota). But during winter, we always want to know how much snow we are going to get. The answer always comes down to ranges
- Less than 1 inch
- 1-2 inches
- 2-4 inches
- 3-5 inches
- 5-8 inches
- 6-10 inches
- 8-15 inches
- Head south
As the totals get larger, the ranges get larger. There is so much room for error due to many variables.
The same thing can be said for estimating network bandwidth requirements for XenApp and XenDesktop. In another blog, I focused on Windows 10 bandwidth for XenDesktop. I want to now look at the same series of tests but for Windows 2016.
I broke the bandwidth estimate down into 3 categories
- VDA Version: Certain releases of XenApp make improvements to the network utilization. Bandwidth tests must account for these changes by looking at different VDA versions.
- Policy: XenApp policies can have a drastic impact on overall bandwidth utilization. With heavier compression (at the expense of CPU), overall bandwidth usage drops. Each test includes a look at WAN and User Experience policies (defined at the end).
- Workload: A user watching videos and browsing Internet content will consume significantly more bandwidth that someone mostly using Office applications. The workloads are broken down across task worker, knowledge worker and power worker.
First, let’s look at the averages for a 60 minute simulation:
By looking at averages, I can make out a noticeable bandwidth reduction for the task worker on a WAN policy between 7.11 and 7.15 releases. To get a better idea on the network bursts, let’s look at the 95th percentile
Again, we see a drop in the task worker with the WAN Policy between 7.11 and 7.15 release. I can also see a trend of dropping bandwidth in the Task worker with the user experience policy and WAN policy.
Let’s now look at application-specific usage:
The bandwidth usage for PowerPoint was surprisingly high. After looking into this, it turns out that there are really 2 numbers for PowerPoint. Presenting and creating. When creating, the bandwidth usage is similar to Word, but when presenting, the bandwidth spikes each time a slide changes.
What about bandwidth usage for video quality?
As expected, as the video quality increase, bandwidth usage also increases. I do find it interesting that with the WAN policy in 7.14, there was a noticeable increase in bandwidth across all video resolutions.Plus, for the user experience policy, there is a noticeable drop in bandwidth usage starting with the 1903 release.
And remember, these tests are simulations. Your results will be different because you are using real users with real workloads with real network congestion.
Note: The naming convention is follows: “Workload – XenApp Policy”
The XenApp policies are
Policy | WAN | UX |
Audio Quality | Low | High |
Desktop wallpaper | Disabled | Allowed |
Dynamic windows preview | Prohibited | Enabled |
Extra color compression | Disabled | Disabled |
Limit video quality | Max 480p/720kbps | Not configured |
Menu animation | Prohibited | Allowed |
Preferred color depth for simple graphics | 16 bits per pixel | 24 bits per pixel |
Target frame rate | 16 fps | 30 fps |
Target minimum frame rate | 8 fps | 10 fps |
Use video codec for compression | Do not use video codec | For the entire screen |
View window contents while dragging | Allowed | Allowed |
Visual quality | Low | High |
Daniel (Follow on Twitter @djfeller)
XenApp/XenDesktop On-Prem Poster
XenApp/XenDesktop Cloud Service Poster
Citrix XenApp and XenDesktop 7.15 VDI Handbook
Hi Daniel,
first of all thanks a lot for the work you put into it. I got a couple of questions 😉
How exactly are you performing the tests? What is your infrastructure composed of? And how much time did you spent on all the tests?
Regards
Christian
LikeLike
An intel nuc with loginvsi to simulate workloads. I’ve automated the entire thing. Once I get through Synergy, I’ll try to post the specifics.
LikeLike
Are u using login vsi to collect the data?
LikeLike
VSI to run tests, using perfmon to gather data
LikeLike
Hi Daniel,
I know this is an old post. I am trying to get a solid grasp on the difference between your 60 minute average and the 95 Percentile.
Looking at the graphs for the Power Worker – ux – 1909
60 minute avg: the value was about 350Kbs for all users.
95 Percentile : the value was about 1,700 Kbs
Are you saying that at the specific time of the 95 Percentile the average utilization for all users in your test was 1,700 Kbs? or are you saying that 95% of all users were below the peak of 1,700 Kbs for the entire test? This peak did NOT occur at the same time for all users.
How many users were you simulating with Login VSI?
Thanks
LikeLike
Think of 95th percentile as the max. If you estimate BW using average, you will run into issues because average eliminates any bursts. If you go with max, you will oversize because max is sporadic. I opted to go with 95th percentile, which means for all of the data points captured, this is the 95th percentile (5% lower than max).
Just simulating 1 user
LikeLike