Could someone please explain to me what is the difference between monthly, weekly and daily average?

From customer point of view, which method would cost you less? Or are they all the same?

I don't know of any provider that takes a monthly average. Most providers take measurements off of their routers ever minute. Generally you can view a daily, weekly and monthly average.

As for what is best for the consumer, the more measurements the router makes, the more accurate your average will be.

It's pretty basic. It's your average traffic for that time period (last day, last 7 days, last 30 days, etc.) The cost would vary depending on your traffic patterns. The most sensible way is to use the monthly average to calculate monthly bandwidth costs. Using the others is risky, you may save money or you may lose money. Say, for instance you average 100kbps total the first 3 weeks of the month. Then the fourth week you use 400kbps. If you were billed based on your last weekly average, you'd be billed for 400kbps. Whereas with the monthly average, you'd be billed for 175kbps.

They all reflect the amount of bandwidth used in a particular period.

For instance, the Daily graph shows your activity in a (roughly) 24 hour period, usually in 5 minute intervals. You can see what your bandwidth looked like in the past 24 hours.

The Weekly graph shows a bunch of those Daily graphs all averaged together to give you a Weekly average. This number should be a close average of the last 7 (or so) daily graphs.

The Monthly graph shows a bunch of those Weekly graphs all averaged together to give you a Monthly average. This is what "most" providers go by as it is "fairly" accurate. It is not "true" or "actual" usage, but it's close.

When you look at it closely, all the charts correspond to each other.

Well if you are asking about using MRTG for percentile pricing - I plead complete ignorance, but there was a good thread on this issue a week or so ago.

However, for some weird reason, statistics was the only area of maths that I didn't fail miserably at. W know that the longer the period over which you sample something out over, the smoother the averages will be in your time series; ie, you will end up with a smoother, flatter graph with less deviation from the mean.

So, with this in mind, you would always want your peaks to be calculated using the longest sampling period (from a customers point of view) and your ISP would want to use the shortest smple period, and that is why I am sure that 5 minute sampling is probably used most of the time.