On 10/6/2005 1:03 PM, Tom DeReggi created:
>> To set the record straight, no peering agreements were violated
>> between L3 and Cogent.
> I heard otherwise, however I can't prove that.

Cogent on their own web site said that agreements were not violated:

"Level 3 terminated its peering with Cogent without cause (as
permitted under its peering agreement with Cogent) even though both
Cogent and Level 3 remained in full compliance with the previously
existing interconnection agreement."

>> There is also no confirmed evidence that L3 is
>> blocking Cogent traffic through Cogent's Verio transit (which Cogent
>> pays $$ for.)
> There was evidence. I wish I saved my traceroutes yesterday.
> To make more clear, Cogent is our backbone.
> When going to www.logmein.com, the last successfull hop was a peer
> labelled similar to verio.cogentco.com, meaning we crossed over to
> Verio's side. (the actual name was more meaningful). Now today, the
> traffic destined for that site stops cold at the first hop from our
> network, meaning it does not get routes from Level3 on where to send the
> data, once we enter Cogent's network.  Unless you are referring that
> Cogent is blocking any advertised route info from Level3, which is
> highly unlikely.  If Level3 was allowing our IPs to go through Verio's
> link, we would receive routes to route our packets in that direction
> across Cogent's network, and packets would travel further into Cogent's
> network (such as to the Verio link). If Cogent blocked traffic to Verio,
> it would most likely block it at the peer, not at the entry to Cogent's
> network from us as their client.

This isn't evidence of blocking on L3's side.  It could be because
Cogent only purchases transit to certain prefixes and L3 isn't one of
them (and Verio is filtering the announcements.)  It could be because
Cogent internally uses traffic engineering to prevent L3 traffic from
reaching them over their Verio transit circuits.  One of the two
scenarios is likely given their peering arrangement with L3.  I didn't
see any table entries on the L3 San Diego looking glass for AS174.  I
saw only one route on their Denver looking glass through AS7018.  Does
that mean that L3 is filtering or that Cogent's announcements aren't
reaching L3 for other reasons?  The former is probably correct, but
that's not something that can be easily demonstrated.  I couldn't find
a looking glass in AS174 which would allow me to see Cogent's tables
from the inside.  Cogent does appear to be announcing their Verio link
to other peers, however.  I see direct announcements for AS174 and an
announcement for Sprint->Verio->Cogent, but not an AT&T->Cogent path.

I think that both carriers are at fault.  Both companies should have
resolved this before it came to reducing connectivity for their
customers. They both should be held accountable by their customers.  I
replied to your original post, Tom, because Cogent made a public
statement which directly contradicted yours and I thought that people
on the list should have a more complete story [1].  You could be
entirely correct about there having been a contract violation.  I am
confident that a considerable amount of money will be wasted trying to
determine that.

I fear that because of the the popularity of this issue it will reach
the ears of the less clueful xEOs at carrier organizations and that
the current SFI structure could be at risk of being 're-evaluated' in
favor of paid interconnection.  Most of the scenarios that I can think
of involving compensation for interconnection lead to higher wholesale
prices of bandwidth and additional overall system complexity.

>> It appears that Cogent is unwilling to use this route
>> because it would force them to pay (Verio) per Mb/s for the
>> information sent to/from L3's network.  The de-peering was consistent
>> with the peering agreement between L3 and Cogent according to
>> http://status.cogentco.com/
> It stated that, but it is not in actuallity.

So why would Cogent lie about something that makes them look bad on
their own public web site?  Many SFI contracts allow for termination
without cause given enough notice and it is reasonable to assume that
this one included that type of language.  According to conjecture on
NANOG, Cogent was given notice >40 days before the disconnect.  In the
absence of more reliable information I don't have any reason to assume

>> Current NANOG consensus (whatever that's worth) is that both companies
>> are equally responsible for correcting their reachability issues, but
>> L3 initiated the de-peering process.
> Agreed.  UNLESS Level3 is actually blocking IPs that were assigned via
> Cogent apposed to just blocking routes or connections. Unfortuneately I
> am not in a possition to prove wether our IPs are blocked because we are
> still single homed with Cogent.  Cogent has so many peers that could
> transmit our data via alternate paths, and the amount of traffic on our
> network going to level 3 is so little, that Cogent would be making a
> poor financial decission not to route our traffic an alternate path
> based on risking that we would switch to a redundant link to Level3. 
> Its not to Cogent's benefit to not route our traffic financial, so it is
> only logical that it is Level3 blocking our IPs.  I was also told Level3
> was blocking our IPs, which is why our IPs could not be re-routed. Sure
> I can't prove this, but its not looking good for level-3.

Since there were no announcements for AS174 present in L3's San Diego
looking glass and there was a route present for them through AS7018
(AT&T) I think that the reachability issues were caused by routing and
not IP blocking, but without direct access to the routing
infrastructure of both carriers, this is difficult to determine.

Generally, SFI contracts do not allow traffic to transit a peer's
network to reach a third-party provider.  While technically traffic
could be allowed to flow from Cogent to PartyA to L3, there is usually
no financial incentive for PartyA to allow this through an SFI and
significant financial disincentive to do so.  Verio provides Cogent
with paid connectivity to certain destinations and theoretically this
transit could be used to reach L3.  Why this isn't happening is a
matter of considerable speculation and in the absence of a statement
from an authoritative source at Cogent will remain so.

>From Cogent's perspective it makes sense (to them anyway) to prevent
traffic from reaching L3 through any means other than the SFI
interconnect(s) because that puts pressure on L3 to bring the SFI up
again.  Cogent has had other SFI circuits disconnected in the past and
there is conjecture that if they don't take a stand, others may follow.



[1] JC Dill recently posted a few more links to other accounts of the

and of course the obligatory slashdot thread:
WISPA Wireless List: wireless@wispa.org


Archives: http://lists.wispa.org/pipermail/wireless/

Reply via email to