I am doing a simulation on ns-2 for a datacenter network and I just realised that I am not sure how exactly I would go to calculate the network throughput.
I know that to calculate throughput over a link I would first trace the last hop link, add the received packet sizes for the correct source-destination pair for the interval I want (1 sec is probably the norm) and then repeat for the next interval and I can plot a graph of throughput vs time but this is just for one link and one pair of nodes. What I am wondering is how do you do this for the whole network? I am thinking that I MUST trace all last hop links to the hosts (i.e. in a fat tree as in my case i would have to monitor the links from hosts to the nodes of the lower level of the pod) fat-tree with k=4 In the case as above I would monitor the links from then black boxes to the first blue circles. But do I add all received packets regardless of source-destination pair? And do I add them for every interval? At the same time I also want to measure queue length (over time). In this case do I monitor the queues of all nodes and average them over the same period of time? Any help/hints/clarification is appreciated. --- This email is free from viruses and malware because avast! Antivirus protection is active. http://www.avast.com
