How to profile a new Application's ICA Bandwdith Usage on a WAN, in the Network Performance / WAN Optimization forum on BrianMadden.com

Hi,

I am in the process of on boarding a new application into an existing PS4.5 environment which will be accessed from the server and delivered to a new site. I am wondering if and how it is possible to profile how much bandwidth a session will place on the link for the new application?

Is there a general rule of thumb that peopple generally use here?

Thanks

There are two things to look at when trying to define the bandwidth used by an ICA application:

a) the "peak" bandwidth usage
b) the "average" bandwidth usage.

The "peak" figure is the minimum per-user bandwidth needed to allow the user / user application to work at a reasonable rate. Yes, if we give the application 10 meg, it WILL use a lot of that. But we may find that it is still usable at only 200k, but we may find that below, say, 50k, it gets a bit sluggish when we do that full-screen repaint.

Of course, only a small number of users will be using this peak value at any given time, but it does indicate the minimum link bandwith that you need (which may be larger than the sum of the averages... see below)

Typically, the peak figure is measured as "the maximum usage in one second across the measured period".


We use the average bandwidthy to work out in a large environment how much bandwidth is needed. If we have 100 users, each needing 10k *average*, then that would mean we need an absolute minimum of 1 meg of bandwidth for that specific application.

Typically, the average is measured by counting the data trasferred over a longer period (1 minute, 5 minutes....) and dividing by the time.


We now have 2 figures, one being the minimum bandwidth defined by the average use of all users, plus a minimum needed to satify the peak usage of a small number of users... from those, you should be able to work out what size of link you need, to provide good user performance.

(of course, none of this takes account of any problms caused by latency... that is a seperate issue)


So, how would you measure this? Well, I would get hold of a Packetshaper, which will be able to give you good peak and average figures, at the aggregate level, the protocol (ICA) level, and (if you set it up) at the per user level. The "protocol level" will give you a great feel of how much bandwidth your ICA traffic is using on your link, but if you are doing testing, you need to home in one one (or more) users, probably by IP, and rund some tests, to get some specific figures.

I'm sure you could get similar data by polling your router every second, and graphing it with MRTG or similar.... (depends if the router can give the "granularity" of data needed for the per-user etc figures)


A rule of thumb? The bandwidth used by any application running via ICA varies depending on the application, but my experience is something along the lines of around 20 to 25k per user average, and peaks of several hundred k.

Of course, any printing (be it via ICA or otherwise) will eat as much bandwidth as is available, and unless you have QoS set up, will probably screw up all of your interactive sessions! (The only effect varying the bandwidth available for printing is to change how long it takes to print! And that is another thing to discuss with your users!!!)


Hope that reply was useful....


Paul Blitz

Thanks for the information Paul - very informative.

Do you know of any tools that can provide the bandwidth usage information? Is this something that Edgesight may be able to assist in?

In my experience, there's only one real tool that can give that level of detail, and that's a Packetshaper.

OK, I'm biassed, I think that the Packetshaper is a great Monotoring & QoS tool, and I'm sure there MUST be other devices (or software) out there.

There's plenty that will give you AVERAGE figures (eg MRTG) but not so much that will give the peak figures.

There's 2 parts to the equation: (a) the device from which you are retrieving the data (eg a router) needs to have the granularity of traffic definition... in your case, just using the client IP would work fine; (b) you need to pull the data every second.

Paul

August 25, 2008 Tom McMeekin wrote:

...how much bandwidth a session will place on the link for the new application?

Tom,

If I understand you correctly, you want to determine how much WAN bandwidth is used? If so, you can either (in the case of telco managed WAN) contact your telco rep to get statistics, or (in the case of in-house managed network) contact your Network Administrator to provide statistics. This can get tricky when attempting to determine a particular processes footprint. I'm not sure if the tools are that granular.

You can look at local server NIC statistics using "Performance logs and Alerts" but again, the Network Interface object doesn't have a Counter to look at a specific process. So by establishing a baseline, you can subtract that bandwidth from what you get when you run 'X' instance/s of the new application - to get your approximate bandwidth utilization.

Hope this helps,