Latest XenApp XenDesktop bandwidth utilization tests

Is it just me, or does it seem like every few releases Citrix finds ways to further reduce bandwidth consumption. When 7.13 came out, there were statements saying bandwidth utilization dropped. And in 7.17, we are hearing similar remarks. If we look at the latest releases, we see the following major improvements to the ICA protocol

• 7.13:
  • Introduce glyph detection/caching algorithm – Beneficial for text-based workloads
  • Improve scroll detection
• 7.17:
  • Update glyph detection/caching algorithm – Beneficial for text-based workloads
• 7.18:
  • H.264 Build-to-Lossless – Beneficial for graphically interactive sessions
  • Progressive Display – Beneficial for low bandwidth and/or high latency connections

I want to see how these improvements impacts network utilization, so I ran my own simulations. I decided to run the same tests against the 7.11, 7.12, 7.13, 7.14, 7.15, 7.16, 7.17, 7.18 and 1808 releases.

Knowledge Worker – User Experience Policy

In the first graph, we have the following:

• Workload: Knowledge Worker
• Policy: Optimize for User Experience
• Test Duration: 60 minutes
• Sample Period: 1 second

I notice the following:

• 7.13: A slight drop in cumulative bandwidth usage. This drop extends through the 7.17 release.
• 7.18: A noticeable drop, most likely due to the H.264 Build-to-Lossless capability with the graphical knowledge worker workload.

Knowledge Worker – WAN Policy

Let’s change the policy and retest

• Workload: Knowledge Worker
• Policy: Optimize for WAN
• Test Duration: 60 minutes
• Sample Period: 1 second

Besides lower numbers, as we would expect with the optimize for WAN policy, we see similar bandwidth reductions in the 7.13 and 7.18 releases.

Knowledge Worker – Comparison

By looking at average bandwidth utilization across the 60 minute test, we can start to see these improvements a little easier.

Task Worker – User Experience Policy

In the first graph, we have the following:

• Workload: Task Worker
• Policy: Optimize for User Experience
• Test Duration: 60 minutes
• Sample Period: 1 second

WHAT???

The 7.17, 7.18 and 1808 releases look crazy. This would be due to the improvements in glyph caching, which is extremely beneficial for text-based workloads, like Excel. I compared the steep increases in the 7.11-7.16 graphs with the actual workload. Turns out these all align with Microsoft Excel. Imagine scrolling through a spreadsheet full of numbers. The glyph detection/caching improvements in 7.17 really shine on this portion of the simulation.

Task Worker – WAN Policy

If we change the policy to WAN and retest

• Workload: Task Worker
• Policy: Optimize for WAN
• Test Duration: 60 minutes
• Sample Period: 1 second

We see similar reductions in the 7.17 release. In addition, we also see the introduction of the glyph caching in the 7.13 release with the optimize for WAN policy.

Task Worker – Comparison

When we change the graph to look at the estimated kilobits per second, the improvements are quite interesting for the Task Worker.

Based on all of these tests, workload changes and policy changes, there are essentially 3 distinct groupings of data based on improvements made to ICA:

1. Group 1: 7.11 and 7.12
2. Group 2: 7.13-7.16
3. Group 3: 7.17-1808

Stay tuned as I plan to keep running these standard tests on upcoming releases.

Daniel (Follow on Twitter @djfeller)
Citrix Workspace Poster
XenApp/XenDesktop On-Prem Poster
XenApp/XenDesktop Cloud Service Poster