Intro

published 1994/5?
analyze routing behavior or path conditions, stability, routing symmetry.
AS synonymous with administrative domain
- BGP scaling lies in the stability of inter-AS routing
  - if routes vary frequently, BGP routers will need to update a lot (bad)

methodology

number of internet sites to run a network probing daemon.
- measure routes with traceroute
two measurement types
- d1: measured all virtual paths between two sites with 1-2 day intervals
- d2: measured between two sites at 60% with 2 hr mean measurement, 40% with mean of 2.75 days
also paired measurements (measure A-->B and B-->A)
time between measurements were IID
- proportion of measurements in a given state equal to the amount of time the internet spends in that particular state
only 37 hosts participated
- miniscule compared to 6.6M hosts and 50k+ NSFNet stub networks
routes are representative because they include a non-negligible fraction of ASes
- internet has only about 1K ASes
samples half of “most important” ASes?

5-8% of traceroutes failed due to NPD.control failing to contact one of the sites
found around 50 persistent routing loops!
- temporary loop is loop resolved in traceroute
  - D1 has two temporary loops, and D2 had 23 temporary loops
erroneous routing
- one traceroute hopped to israel instead of going to london on a transatlantic journey
fluttering (rapidly oscillating routing)
- occurred most frequently between wustl and umann
  - 17 vs 29 hops - this splitting is technically allowed but probably not appropriate. Hard to get good routing metrics
infrastructure failures:
- traceroutes failed a number of times
  - d1: 99.7-99.9 success rate
  - d2: 99.4-99.6 success rate
unreachable hops
- internet is > 30 hops, need higher TTLs
- distance not always correlated to hops
  - 3/5 hops for 1500km/2000km distances
  - 11 hops between MIT/harvard both directions
outages followed typical working hour phenomenon
- least in 01:00-02:00
- most in 15:00-16:00

In general, Internet paths are strongly dominated by a single route, but,as with many aspects of Internet behavior, we also find significant site-to-site variation
route changes exist within 10 minute intervals
25 instances across 1302 10-min observations

49% of measurements observed an asymmetric path that visited at least one different city
consider AS instead of city - still 30% of traffic followed asymmetric paths