Paul, This is the scenario. Peer B is send lots of outbound to Peer A. Peer A depeers Peer (well former Peer) B. Why? Well, Peer A is having ratio problems with other Peers C-F. Keep reading... After depeering, some of (now former) Peer B's outbound traffic to Peer A will now flow over links from Peer B to Peers C-F, before finally terminating at Peer A. Peer A sees their ratios with Peers C- F improve. This is a proven maneuver and Cogent is not the first to do it. Of course, it gets more complex with multihoming and the assumptions of a meshy enough connectivity to ensure this will happen. This is better explained with a whiteboard. That full explanation was missing from the writeup that is posted (and I'll allow it to stay up for now), because that report was aimed at folks who may not be fully conversant in peering - financial professionals. BTW, thanks for dropping me an email to ask me about it, before posted to NANOG. As far as reachability from one provider to another - I've heard that one can make routing changes quickly and easily on this crazy Internet thing. Perhaps in the 24 hours since I wrote that, a few changes occurred? Dan On Sep 28, 2007, at 6:00 PM, Paul Vixie wrote:
at <http://www.e-gerbil.net/cogent-t1r> there is a plain text document with the following HTTP headers:
Date: Fri, 28 Sep 2007 21:56:34 GMT Server: Apache/2.2.3 (Unix) PHP/5.2.3 Last-Modified: Fri, 28 Sep 2007 19:15:53 GMT ETag: "92c1e1-a85-43b36ea5bcc40" Content-Length: 2693 Content-Type: text/plain
the plain text title is:
Cogent shows hypocrisy with de-peering policy
the plain text authorship is ascribed to:
Dan Golding
the first plain text assertion that caught my eye was:
Cogent, has, in fact, de-peered other Internet networks in the last 24 hours, including content-delivery network Limelight Networks and wholesale transit provider nLayer Communications, along with several European networks.
since i appear to be reaching the aforementioned web server by a path that includes cogent-to-nlayer, i think this part of the plain text is inaccurate.
traceroute to www.e-gerbil.net (69.31.1.2), 64 hops max, 52 byte packets 1 rc-main.f1.sql1.isc.org (204.152.187.254) 0.336 ms 2 149.20.48.65 (149.20.48.65) 0.509 ms 3 gig-0-1-0-606.r2.sfo2.isc.org (149.20.65.3) 1.163 ms 4 g0-8.core02.sfo01.atlas.cogentco.com (154.54.11.177) 2.757 ms 5 t4-2.mpd01.sfo01.atlas.cogentco.com (154.54.2.89) 2.958 ms 6 g3-0-0.core02.sfo01.atlas.cogentco.com (154.54.3.117) 2.525 ms 7 p6-0.core01.sjc04.atlas.cogentco.com (66.28.4.234) 4.183 ms 8 g3-3.ar1.pao1.us.nlayer.net (69.22.153.21) 2.637 ms 9 ge-2-1-1.cr1.sfo1.us.nlayer.net (69.22.143.161) 3.806 ms 10 so-0-2-0.cr1.ord1.us.nlayer.net (69.22.142.77) 69.022 ms 11 60.po1.ar1.ord1.us.nlayer.net (69.31.111.130) 69.491 ms 12 0.tge4-4.ar1.iad1.us.nlayer.net (69.22.142.113) 81.580 ms ...
the second plain text assertion which caught my eye was:
Why is this happening? There are a few possibilities. First, Cogent may simply want revenue from the networks it has de-peered, in the form of Internet transit. Of course, few de-peered networks are willing to fork over cash to those that have rejected them. Another possibility is that Cogent is seeing threats from other peers regarding its heavy outbound ratios, and it seeks to disconnect Limelight and other content-heavy peers to help balance those ratios out.
this makes no sense, since dan golding would know that cogent's other peers would not be seeing traffic via cogent from the allegedly de-peered peers.
so, i think the document is a hoax of some kind. (i saw it mentioned here.)