did AS174 and AS4134 de-peer?

Jim Cowie cowie at renesys.com
Wed Mar 7 18:06:12 CST 2012


On Wed, Mar 7, 2012 at 6:33 PM, Patrick W. Gilmore <patrick at ianai.net>wrote:

> On Mar 7, 2012, at 18:29 , Nick Hilliard wrote:
> >  On 7 Mar 2012, at 23:19, Darius Jahandarie <djahandarie at gmail.com>
> wrote:
> >> On Wed, Mar 7, 2012 at 17:55, Greg Chalmers <gchalmers at gmail.com>
> wrote:
> >>>
> >>> Isn't this journalism a bit yellow? No facts / based on speculation..
> >>>
> >>> - Greg
> >>
> >> Now all they need to do is link back to this NANOG thread as a source.
> >
> > That would be very irresponsible. Otoh, if someone updated the tier1
> network page on Wikipedia first...
>
> There is no change to the list.  Cogent still does not have transit.
>  Cogent sees CT through Sprint (a peer) because CT pays Sprint for transit.
>
> OTOH, Jim did say in his blog post: "This disconnection will increase
> China Telecom's transit costs...."  This assumes facts not in evidence,
> namely that the CT <-> Sprint pipes were not full before the de-peering
> incident.
>
>
Heh.    I think Doug was pretty clear in his summary of the observed facts,
at least.  There was a healthy, longstanding routing adjacency, observed by
all.  Right sharp at the top of the hour (10:00pm in China, 9:00am Eastern
time), that connection disappears from global view.  Afterward, the
percentage of the Renesys peer base that likes transit paths to CT through
Sprint ticks up modestly.

The real story there is hidden in that traceroute latency plot.  Look how
neatly it bifurcates post-event into paths through Sprint and paths through
Level3.  Notice that paths through Level3 tend to have slightly lower
latencies and significantly less volatility.  Infer what you will about the
congestion on the Sprint-CT pipe.

As a meta-comment: this "Quick Look" style of blog is an experiment we're
trying, based on feedback that the community wanted to hear about more of
these little events as they happen.  In a Quick Look, we're giving the
facts as they are known from initial measurement, and a very quick summary
of our preliminary analysis of the incident.   Then we throw the topic open
to comments from those who might have the clues to the rest of the story ...

cheers,   --jim


More information about the NANOG mailing list