From: Susumu Harada (harada@cs.washington.edu)
Date: Wed Nov 03 2004 - 01:41:37 PST
"Quantifying the Causes of Path Inflation"
N. Spring, R. Mahajan, and T. Anderson
This paper sought to measure the severity and identify the causes of
path inflation through detailed analysis of network traces gathered over
several days and spanning a global scale. They classify the possible
causes into a combination of topology and policy along each of the three
levels of interconnectivity scope, namely intra-domain, peering, and
inter-domain routing.
They discovered that although over half of the path inflation is
attributed to inter-domain routing policies, that this is due not to
intentional network engineering on the part of the ISPs but rather due to
the limitation in the functionality of BGP (Border Gate Patrol). The
authors suggest that one way to augment BGP to help alleviate path
inflation is to share geographic information about the peering nodes of
each ISPs, allowing the ISPs to make better routing decisions for choosing
appropriate exit points.
They also observed that the ISPs appear to be operating in collaborative
fashion contrary to popular belief, based on the fact that the ISPs
appear to be performing some kind of load balancing behavior over their
different exit points, as was seen in the case of San Francisco traffic
being routed all the way to Seattle. This also led to their conclusion
that load balancing policy that is oblivious to the network topology can
significantly worsen the path inflation.
The paper was strong with its amount of data to back up their
observations, but I had a concern regarding the manner in which they
calculated the hypothetical baseline against which they compared their
empirical measurements. Their assumptions about the black box
network components, especially regarding the inner workings of an ISP
network, seemed too oversimplified and I would have liked to see more
concrete validation of their assumptions.
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 01:41:38 PST