> Sorry to be so late coming into this, but we did some work on caching > effectiveness at PEs, see section 5 of our technical report: > > http://fmdb.cs.ucla.edu/Treports/080004.pdf
There is some terminology used in section 5 that I'm not in tune with, but I still think I got the gist of it. If I read it correctly, the simulations tried to use "typical" traffic patterns to validate the behaviour, with a relatively limited population being served by the network at the sampling points. There are two potential problems with this when one look at it on a wider scale: 1) From an operational perspective, one is much more interested in robust behaviour under *any* circumstances. So, in addition to looking at "typical" traffic traces, one should also study the worst-case behaviour. 2) My guess is that if one tries to employ caching solutions when one serve ever larger populations (as in the current Internet, when e.g. performing service in large ISPs networks), caching solutions will come under significantly more strain, since my expectation is that the forwarding will encounter ever more entropy/randomness, decreasing the utility of caching solutions. I'm old enough to remember the first-generation "route cache" used on Cisco AGS+ routers, and the initial design's reliance on "typical traffic patterns" -- if I recall correctly it inserted host entries in the destination forwarding cache. It performed reasonably well under normal circumstances, but fell over when the first user in rather quick succession tried to sequentially "ping" all the addresses on the Internet... That lead (via several intermediate steps) down the path to the current situation where Internet routing gear typically does not operate with forwarding caches, but instead with fully populated FIBs for use by the forwarding logic. Regards, - Håvard _______________________________________________ GROW mailing list [email protected] https://www.ietf.org/mailman/listinfo/grow
