Review: Quantifying the Causes of Path Inflation

The paper analyzes the causes of path inflation by conducting a
trace-driven study of 65 ISPs. They characterize and describe the root
causes of path inflation as the following 6:
1. Intra-domain factors: effects on paths from a source in one domain to
a destination in the same domain (ISP)
    a. Impact of topology: topological inflation among tier-1 ISPs was
higher compared to the other ISPs.The authors also observed to
significant correlation between topological inflation and ISPs
geographic presence.
    b. Impact of routing policy: The authors inferred routing policies
by assuming that the routing is weighted shortest path with some unknown
set of edge weights, and they then determined the weights that satisfied
the paths observed. Once again, there were differences between tier-1
and other networks, but not between ISPs in different parts of the
world. Furthermore, an important observation was that intra-domain
traffic engineering is consistent with latency-sensitive routing, that
is, ISPs try to minimize the latency while also balancing traffic across
backbone links.
2. Peering factors: effects that occur when the source and destination
are connected to adjacent ISPs.
    a. Impact of peering topology: it does not inflate paths significantly.
    b. Impact of peering policy: The results show that these often cause
path inflation (about 30% of paths experience inflation), and sometimes
lead to highly inflated paths. This means that they consume more network
resources than an optimal exit policy (a policy that chooses the peer
point closest to the source and destination). Furthermore, the authors
observed that inflation quickly decreases as the number of peering
points increases because the traffic now has more links over which it
can be routed.
3. Inter-domain factors: effects long paths that traverse multiple ISPs.
    a. Impact of topology (AS-graph): The authors conclude that crossing
multiple ISPs does not appear to contribute significant additional path
inflation.
    b. Impact of inter-domain routing policies: The paper says that
inter-domain routing in the current Internet has significant impact on
path inflation with more that half of the paths being longer that the
shortest path. The most important contributing factor is not the
policies arising out of commercial concern, but rather the use of
shortest AS-path as the default distance metric.
The paper concludes that the two biggest factors of path inflation are
inter-domain and peering policy. Furthermore, the authors believer that
path selection can be significantly improved and path inflation reduced
if effective mechanisms were made available to ISPs to better engineer
peering point and AS-path selection, for example appending geographic
coordinates to route advertisements in BGP. The authors claim this
improvement is possible because intra-domain routing is already latency
sensitive and ISPs already cooperate amongst each other to engineer
better peering exits, which suggests that ISPs are also willing to work
with each other to achieve low latency routing.

The paper is well written. I like how the authors have clearly and
neatly characterized the factors they studied, and presented the results
in a concise and short manner.
Received on Wed Nov 15 2006 - 23:04:47 EST