Network Probe Daemon -- runs on sites around the Internet. These sites are not "special" per se. The daemon take authenticated requests to run traceroute, tcpdump and source/sink a probe.
A large number of sites in US, Asia and Europe. There is a n-squared effect when sites are added.
Two runs were done. The first one involved 6459 traceroutes sampled every 1-2 days in the last two months of 1994. The second one was done the last two months of 1995. There were 35,109 traceroutes with two rates (mean 2 hours for 60% of the paths and 40% with a mean of 2.75 days)
85 ASs samples (about 8% of the total active ASs) (Of course, not all ASs are equal. Since some ASs are bigger than others, about 50% of the ASs that are covered.)
Total user visable pathologies did get worse.
A route that is persistent is also prevalent. However, a prevalent route may not be persistent.
Internet paths are strongly dominated by a single route.
More than 2/3s of routes persist for more than a day.
Is the path from A->B the same as B->A.
About 30% of paths in 1994 were asymmetric.
About 50% of paths in 1995 were asymmetric.
The magnitude of the asymmetries were were using just one city, though many were two.
Yakov commented that traceroute is not an accurate tool. He wonders about the error in the data. --- It could be large. The data has been statistically analyzed to remove as much error as possible, but this is not an attempt to characterize the errors in traceroute, per se.
Is the number of hops growning and reducing? In the first study, there were no sites beyond 30 hops and in the second, there were definately some that were 30 or greater.
It appears that most of the sites in the sampling pool were academic. Is the sample pool valid? Yes, the sample pool is valid since anything specific to the site where statisically eliminated.
Why consider cities as significant granularity? It was a way to represent specific geographical distances. It might not be a great approach.
Since most sites were academic, what impact did the shutdown of NSFNET have? Surprisingly, very little.
Enke is not surprised that there is more asymmetric routes. Enke thinks that as more new BGP features are available and used, this may be moderated more.
Enke says that the data presented did show a large number of routing instabilities. He is hoping that Vern will rerun this again in 1996. Vern said that he didn't know if he will be able to do that. However, Vern does suggest that some kind of mechanism be developed to make gathering this data more routine.