Hi Devang,
Without getting into the complexity of the actual formula (which is not really all the complex, but looks that way), here is a brief explanation:
- it is important to use an average queue depth value for RED instead of an instantaneous value because this allows you to provide a smoother response. Imagine a link on which you started dropping packets the moment your instantaneous queue depth reached 75% (an example). If you did that, you would end up over-reacting to this bit of congestion. It may happen that the queue is totally empty a second later but you have randomly dropped some packets and caused some flows to re-transmit their packets. If an average value is used, the average will increase on congestion but it will increase at a lower rate
- the reason you are advised not to change the constant is because it has been derived by Cisco based on extensive RED testing of real flows. Therefore, Cisco feels that this value is quite optimum and should not be changed unless you have intimate knowledge of the sort of flows you will see on your network ...
As for the formula, which is:
average = (old_average * (1-1/2^x)) + (current_queue_size * 1/2^x)
The above can be simplified to
average = old_average * 0.998 + current_queue_size * 0.002
The above was done simply by plugging in a value of 9 for the exponential weighting constant (x). So you see, it's not that complicated at all...
Hope that helps - pls rate the post if it does.
Paresh
