Check your AS: Renesys Blackout Report Released

Mon Nov 24 19:16:43 UTC 2003

Folx,

Hot off the presses, from the people who brought you the excellent (and
fun!) reports on the effects of worms on routing instabilities, how the
Internet fared on Sept 11, 2001, and other fine topics of interest to the
operator community, comes a new report:

"Impact of the 2003 Blackouts on Internet Communications"
available now at:
http://www.renesys.com/news/index.html
<a href="http://www.renesys.com/news/index.html">here</a>

It is an attempt to do a thorough, retrospective analysis of the impact of
the power outages from a purely routing perspective.  We tried to be quite
rigorous in our methodology and careful in our inferences.  However, we
came to what may be an unpopular conclusion:  the Internet fared worse
than others have previously reported.  The main difference in our
conclusions lies in different measurement strategies (core to core layer
3-4 monitoring versus global BGP routing monitoring).  Read the paper for
more information.

We also hoped to produce a definitive analysis of the network (routing, 
BGP) impact of the power outages so that others can compare future events.  

We're particularly interested in feedback from operators with assets in 
the affected regions of the US, Canada and Italy (see Appendix B for a 
good comparison of the Sept 28 Italy Blackout with the Aug 14 US 
Blackout).  

A few specific ASes are mentioned in the report. We would love to hear
feedback from those ASes or others who were affected to learn more about
the backstory behind the outage.  If your prefixes stayed up, why?  If
some went down and some didn't, what caused that?  Did your upstreams and
peers stay up?  Were local power outages at routers the primary cause of
outages, or did other factors enter into the equation?  We saw one AS with 
nine (9!) upstream ASes lose all of it's prefixes.  Could it be that 
someone with 9 upstream adjacencies didn't have reliable power? 

These questions, plus a general discussion of Internet edge reliability 
(power and interconnectedness) seem on-topic for the list.

Of course, we read nanog :-), so we'd love to see those stories discussed
here in a context that would help all of us understand the causes and
mitigation strategies better, but private mail will also be gratefully
accepted.  If you don't ever want us to mention your name in public, be
sure to let us know.

Todd Underwood
todd at renesys.com