I work in a high school, and want to do a performance test of our wireless network. We have some 700 students, each with at least 1 wireless device. Assuming they will typically only use one device for any sort of high-throughput activity, that is about 700 devices. There are plenty of WAPs distributed around the school; roughly one per 25 students. The WAPs are all dual band 802.11n, and use a wireless controller. My idea is to have all users simultaneously download a large(ish) file, and record throughput both at the server and client level. Any considerations how I should serve such a file to stop disk IO being a bottleneck? What about the ethernet to the server being a bottleneck. Assuming each client having about 10Mbps, this would require the server to have some 7Gbps of network goodness available. Is this a valid concern? Are there any client-side tools that can synchronise the transfers and automatically report the speeds the client saw?
Alternately, if this is not a valid testing methodology, how would I go about doing a wireless performance test?
As stated in my comment above I couldn’t see a reasonable reason why you would want to do this unless you had a problem, and if you had a problem I would look at other troubleshooting before doing something like this.
If you do your proposed test and get each client to download your test file I feel that you are going to saturate your network entirely.
You have so many variables with testing this as well. Each client is going to be a different distance away from the AP. They will all be using different chipsets, (This WILL affect performance). Even if you automate this, you will get different results (very subtle) but a result from a client that has to take 2 hops to get to your server to a client that takes 4. As far as your lest part of the question goes. Has anyone complained?
If not don’t fix something that isn’t broke.
I can't help thinking this test is invalid as a test of how clients will behave in actual use, unless "everyone downloading massive files at the same time" is the actual usage scenario. And if that's the case, then performance will probably be problematic no matter what you do; this is one of the use cases where wired is better than wireless.
You could write a script and run it on a representative sample of machines that mimicked a typical "log in, open a document or three, check a website then logout" scenario, run it in one localised area at a time (to ensure you're testing the same group of APs) and average it out and get (imo) more realistic results.