===================================================
From takhoa@rice.edu  Wed Feb 11 19:02:32 2004
===================================================

This is a very good paper as it characterizes different types of routing
instability in the network.  Particularly, the paper does a good job of
pointing out how different types of routing instability events contribute
to the overall problem of instability.  This paper gives network
implementors good ideas of what to focus on in order to improve the
performance of the Internet.  The authors also provide good explanations
of how they come up with certain methodologies and calculations.  They
also offer good suggestions into possible causes of the excessive routing
packets.

Some of the points in the paper are not very well discussed, however.  For
example, the authors showed that the Internet routing tables are dominated
by a few ISPs.  However, they do not discuss the percentage of Internet
traffics being served by these ISPs.  It might be the case that the
routing tables are shared proportionally to the amount of traffic each ISP
serves.


===================================================
From dushu@cs.rice.edu  Thu Feb 12 15:18:14 2004
===================================================


The authors studied the instability of routing in the Internet 
by observing, and thereafter analyzing, the trace logs from some 
backbone route servers in the Internet. They found that a major 
chunk of the routing management traffic goes to the pathological 
information, that is, the redundant routing update information, 
although the authors didn't show the convincing evidences that 
these redundant routing updates had significantly decreased 
routing performance in the Internet. So I think it's
an interesting work, but its impact needs to be tested or I hope 
I can hear it today since this was a paper in 97.

===================================================
From santa@cs.rice.edu  Thu Feb 12 16:00:54 2004
===================================================

Internet Routing Instability

This paper examines internet routing instability by measuring it and coming up
with possible explanations. High levels of instability leads to packet loss,
latency and time to convergence, giving bad network performance. The magnitude
of routing updates is measured to be many orders of magnitude more than
reasonable. This unstability typically occurs due to router minconfigurations,
transient failures and software bugs.
The methodology they used was to instrument the route server which has peering
relationship with majority of ASes at the public exchange points. Routing
information is updated betweeb BGP peers by announcements and withdrawals. They
define many of these routing update instabilities by Withdrawal/Announcements
difference/duplication.  The instability is well-distributed over destination
prefixes and peer routers. But, analysis showed that most of the pathological
routing was caused by small service providers - possibly due to more number of
router misconfigurations.
Investigating the possible causes of thes instability, the authors found that
many router vendors have implemented stateless BGP leading to increased number
of updates. Drift in clock synchronization also seems to account for some of the
updates, as router interface cards are very sensistive.  Unjittered interval
timer leads to overloading of the CPUs of routers making the routers stop
processing future packets for some time, giving to frequent transient loss. The
interfacing between EGP and IGP also seems to be a cause of concern.