apologies for latency On Wed, 23 Apr 2003, Matt Larson wrote:

I'm looking for statistics on the composition of Internet backbone traffic broken down by application protocol. Ideally, I'd be interested in a list showing the relative percentage of packets by TCP and UDP destination port of some large sample of Internet traffic. But I'll take any statistics related to application protocol traffic distribution that I can get.

If anyone has information that they would be willing to share publicly or privately, I would be grateful for it. Alternately, if anyone is aware of such information publicly available, I would be grateful for pointers to that, too.

When in doubt, if you are looking for statistics about or on the Internet start with www.caida.org. caida website is undergoing restructuring for easier navigation (yeah we'll be done any year now) so in the meantime let me make this easier on you also allow me to lower your standards (but raise your future expectations) because we are in the process of setting up a better system for regularly capturing and maintaining this type of data. (blocked on hardware and software and some cool methodology in development to do high speed flow estimation/analysis) anyway so you may want to try the 3 links under http://www.caida.org/analysis/workload/byapplication/ most recent in webpage form (not so recent) for high-capacity backbone (san jose, california): http://www.caida.org/analysis/workload/byapplication/oc48/20020305/apps_perc... 20305/20020305_b.xml older: http://www.caida.org/analysis/workload/byapplication/oc48/20011029/apps_2001... tabular view of same: http://www.caida.org/analysis/workload/byapplication/oc48/20011029/apps_perc... finally it's academic-ish locations (but hey that's where all future application types start) and older data (1998-2001, we're trying to update it w newer data this spring, stay tuned) but try figure 7 of http://www.caida.org/outreach/papers/2003/nlanr/ (`Longitudinal study of Internet traffic from 1998-2001: a view from 20 high performance sites') sorry it's suboptimal but so is the current ability to measure this sort of thing at a large and arguably representative scale but you knew that already k