How to find the effective bandwidth correctly using statistics?

I have done an experiment to find the effective n/w bandwidth.

The data I got in kbps is

223, 221, 510, 220, 471, 229, 222, 221, 220, 221

How can I find the effective bandwidth? Averaging gives 275.8. But if I have done only first 4 rounds then the average is 293.5. How can I find out a more reasonable value as the effective bandwidth. Or is averaging the correct way of doing this?

• n/w = network? Isn't packet-transmission data usually long-tailed? – cardinal Apr 10 '11 at 15:31
• n/w = Network. Sorry for using the abbreviation. I have transmitted 1MB of data to the peer system. like that I computed the bandwidth – pkvprakash Apr 15 '11 at 4:08

1 Answer

From what I can understand, I think you have the following options:

1. Sample more! n = 10 is hardly enough for drawing conclusions
2. If you don't/can't do "enough" sampling, you can always try to do some Monte Carlo type study with bootstrapping
• So averaging is the most recommended way? Can I conclude like that? – pkvprakash Apr 15 '11 at 4:03
• I am not sure what you want to do would be an accurate measure of the bandwidth. The bandwidth represents how much COULD be transmitted down the connection and not how much actually gets transmitted. If you want to get a measure of average tranfer speed then yes, averaging would be the way to go. Regardless, I would prefer to supply mean, median and a std deviation together, then you summarize your data better – posdef Apr 15 '11 at 7:27