Forum

Hi folks,
I'm interested in doing a little test to satisfy my curiosity. I'm going to use Windows Performance Monitor (perfmon.exe) to analyze the impact that different authentication and encryption methods might have on a given PC, either on memory, CPU overhead, etc. I'm going to examine all the authentication methods available, including Open, WEP, WPA, etc., etc.

Can one on you really savvy gurus give me an idea of which performance objects would be most impacted by the use of encryption that I should include in my test? I'm thinking CPU or memory usage would be the most obvious factors impacted? I would welcome your suggestions.

Suggestions? Comments?

Thank you,
Craig

Hi Craig,

Some of your questions may be answered here:
http://www.microsoft.com/whdc/device/network/NetAdapters-Drvs.mspx

tom

Make sure you track kernel mode processes (proviledged processor time) as this is where drivers reside.

Let us know your findings. A spreadsheet with the different encryption algorithms listed and the average processor utilization during an exact window would be helpful. You will probably need to script the transfer of data using WSH so that you can set a write a start time to a variable, send the data, write an end time to a variable. This will allow you to track the amount of time it took to transfer the data.

Next, you can load your perfmon logs and narrow the view window to exactly the start and end times for monitoring. This can show you all your stats in that window only.

Sounds like an intersting project. Let us know what you find.

BTW - It is very difficult to compare apples-to-apples in these situations. You will probably want to unload all unneeded services before doing the test and reboot right before each new analysis window to make things as clean as possible. If the results show only slight variances, it may not be very meaningful... so make sure you transfer enough data to monitor over a long enough period of time to get an acceptable margin of error.

Just a thought...

Craig,

How are you going to determine whether any given test result is an
affect of the network or of the pc? You may be in the middle of a
stateless network transaction, udp, and some OS housekeeping
task may kick in and affect how your network driver can get to
the shared resource of main memory, and/or the hard disk.

Encrption will affect things on a large transaction, but authentication
should only affect initial BSS association metrics and then be static.

Tom

Thank you all. :-)

Tom (dumbmick): To answer your question, "How are you going to determine whether any given test result is an affect of the network or of the pc?"

I'm not sure, other than to just try and do as uniform of test as possible, i.e. conducting my test(s) the same time of day, etc., etc., etc. I'll be conducting these tests at my home using a Cisco Aironet 1232AG 802.11a/b/g Wireless Access Point connected to cable broadband. I'm thinking of establishing a connection to the AP, and then perhaps downloading a large file (service pack, etc.), and/or doing some FTP transfers, etc. I'll use perfmon to track all of the impact on the PC (and maybe I can track the throughput as well).

Craig

Hi Craig,

Your methodology of running the tests at a particular time may not affect their repeatability or reliability. The reason is that you are using a non-specialized computer, a pc, to do this testing. A pc is like a sedan and functions very well as such. Put it on a drag strip, and the sedan is toast. What you need is either a specialized software specifically written to test networks devices, or a specialized computer such as from Azimuth or Agilent, etc. . These of course cost mad amounts of money, and are out of the question as a possible test tool.

What your trying to do, however, is very worthy in terms of understanding networks and how various configurations affect their performance, and I commend you for the idea. I'd love to see the outcome from your tests and would be happy to help you with a test plan if you would like, but be forewarned that the two main R's (reliability and repeatability) may be elusive in such a test setup.

Regards,
Tom

Thank you, Tom (dumbmick)...that's very nice of you to offer help. :) I certainly don't want to burden anyone with my little experiment, but if you have suggestions I am happy to try and implement them. I think my little experiment will take a week or two to complete, just due to my very busy life.

Before I start, I think I'll post a message to a Microsoft newsgroup, and ask a similar question; what Windows processes, that can be measured using "perfmon", would be most indicative of wireless network traffic. I'll capture the results using a custom perfmon Counter Log filter, import the data to a spreadsheet, and post it to this newsgroup so someone can help me interpret the results. :D

The one thing I'm really curious about is to see the impact, if any, that different WEP & WPA key lengths have on performance overhead.

Thank you all for your input...stay tuned...

Craig

Those encryption methods will have an impact, but your neighbors
AP, and it's ensuing interference, will probably have a higher impact.

Although old and overkill for what you want to do, check this out:
http://www.reed-electronics.com/tmworld/article/CA457477.html

Ciao,
tom