Hi, I am trying to calculate trhoughput and queue sizes in different levels of a fat tree and this is how I went about it.
I had /trace-queue/ from all edge/tor nodes to the hosts and used those files to calculate throughput on the edge-host level by calculating the received packets every 10ms on each host and added them all up. For queues I used /monitor-queue///at an interval of 1ms in the pod level (between edge/tor and agraggation nodes). Actually I had only monitored from edge to aggregation node but I should monitor both ways right? To calculate the queue size I was using [expr $parrivals_ - $pdepartures_ - $pdrops_], is it correct? So one of the reasons I did was also because I wanted to limit the space I have to use - so I was thinking that I should probably record both throughput and queue size on every level - host->edge, edge->aggragation, aggregation->core - but while I am pretty sure that I can use the addition of thrroughputs at the corresponding level, can I use the average at every interval of all the queues at the corresponding level? Also, I am using trace-queue on the lower level but monitor-queue on the pod level? Can I calculate throughput using monitor-queue as well, it would probably really limit the disk usage and I am guessing less writings to files should also improve performance but I am not sure something might be lost if I don't have the trace. Andreas Andreou --- This email is free from viruses and malware because avast! Antivirus protection is active. http://www.avast.com
