What’s the best configuration for WAN simulation of remoting protocols?

There are boatloads of WAN simulators out there and I have used most of them.  The one we are now using here at Cisco for VDI design work is from Apposite (www.apposite-tech.com/.../overview.html).    The key is to get one that has large enough buffers to accurately induce delay and loss based on known values or variance.  You do not want one that drops packets just because of small buffers. This skews your results and end-user experience reporting.  Also, you should test with various delay values (T1@80ms, 120ms, 200ms, 3M@80ms, 120ms and so on).  Loss values can be fixed or variable and usually range from .005% loss to a very very crappy 5% loss (common in links from hotels or remote sites in developing countries).

I have a Cisco Validated Design (CVD) guide being released at the end of this month that leverages a WAN simulator for RDP and PCoIP comparisons with VMware View 4.  The testing will compare RDP and PCoIP with direct mode and tunneled mode as well as MMR and USB redirection flows.  Let me know if you need help. :-)

Ditto to what eyepv6 said, we are also using the Apposite. I like its ability to simulate variable latency and add traffic load to the pipe. Its real-time by the second reporting and playback is also cool. We support a site in India and this device has been crucial to proving our case that although forms of SBC work, there are ones that handle the 220 to 300ms latency better than others.

We compared PCoIP and HDX over WAN too. We don't use any tools for that. What we have done? We have simulated a normal user, who used PowerPoint, Word, Internet, Flash, ... I think to feel the real User Experience is better than any statistics. ;)

The decisive factor was also a test over SSL, which is not supported with PCoIP - today.

I am a strong believer that if you want to simulate the real world WAN you want to make the delay variable.  By that I don't mean testing with different latency values, but use settings that dynamically change the latency, often called "jitter" or "variable delay".  

When it comes to doing so, you may find different kinds of variable delay.  "Gausian" is like random noise.  "Markovian", if available, would be better for what you want.  It works kind-of like this.  Randomly delay a packet, if the delay is big, it increases the odds of the next packet having a big delay.  This is more like the behavior of a congested link or routere buffer.

If the tool doesn't provide variable delay, you can cause the same effect by variably adding additional traffic sources on and off through the same link.

Brian,

Why not use Tim Mangans TMNetSim Network Simulator ?

The description sounds just about right :)

"PC Platform software to simulate a bad WAN on a PC"?

http://tmurgent.com/Tools.aspx

Depending on your available resources, WANem is a free OSS project that I've used for WAN emulation with good results:

http://wanem.sourceforge.net/

virtualizationreview.com/.../handy-tool-for-vdi-pilots.aspx

Kind of off topic (and common sense), we also used the WAN simulator to illustrate the real-time bandwidth utilization by the second. This was very important because often times WAN providers will give you utilization reports that span a 24 hour (or longer) period. In the case of real-time technologies like VoIP and SBC, these reports paint a false perception. To the layman, these reports may show a T1 / E1 with only 40% utilization however during a VMware View / ICA user session there may be millisecond / second peaks depending on what the user is doing. These peaks often translate to a text delay or mouse freeze for the UX.

Also, another off topic finding is that we saw that PCoIP, over the WAN, did not exhibit as many high peaks but instead maintained steady utilization.

It's important to factor in content type and impact on bandwidth. I would suggest try multimedia content like flash and windows media player, and then try regular apps. This will expose that PCoIP with UDP works well on latent connections at high bandwidth for multimedia, but when you go to regular apps it's a different story, especially on bandwidth consumption. That's what my testing has shown to date. I'm not about to upgrade my WAN to support PCoIP so this stuff matters.

As for latency, I generally use 80ms for US based testing. 130ms US-EMEA or EMEA to Asia. I use  240ms-360ms to simulate US to Asia delays depending on if I go over the Atlantic or Pacific.

As for bandwidth itself, I typically use T1 i.e  1.5M, 2M-3M (good test for the non multi media apps) I have many links in Asia like this, 10M for my major pipes to EMEA. T3 i.e 45M may also be worth it but it's a edge case IMO.

@Kimmo

Thanks for the thought.  I think Brian has access to better quality tools than mine.  The best thing about TMNetSim is that it is free ;)

@Tim Mangan

Not to mention I gave a presentation at BriForum on the topic a couple years ago, using your tool and a couple others Tim! I am almost certain that Brian has access to the BriForum videos. If not he can always buy the BriForum DVD from TechTarget.

One scenario that should be tested is the branch office environment with a few dozen users.  For example, a branch office needs to access VDI desktops at headquarters.  They are connecting over a private VPN to the headquarters on a T1.  

Let’s say there are 30 users in the office, that means about 50Kbps for each user on the T1.  

Since not all users will be working at exactly the same time, let’s give each session 100Kb/s.  

Average latency can be set at 70ms (the office is across the country).  Typical application usage would be MS Outlook, PDF scrolling, and a graphical PowerPoint presentation.

It would be helpful for us to see how PCoIP performs in this scenario, and compare its performance to Microsoft RDP7, Ericom Blaze, Quest EOP, Citrix ICA, and WYSE VDA.

Hey Brian,

I use WANem in my demo lab as documented here: www.jhouseconsulting.com/.../wan-emulation-for-vdi-and-sbc-testing

As pointed out by Tim, for real testing you also want to be able to introduce jitter, out of sequence packets, etc, so you would need to run WANem in router mode and not bridge mode as I do.

Having said that I'm sure you can get your hands on either of these awesome commercial products...

www.networknightmare.com

http://www.shunra.com/

Cheers,

Jeremy.