Operational impact of depeering

Mon Oct 10 13:46:21 UTC 2005

On Oct 10, 2005, at 9:28 AM, Michael.Dillon at btradianz.com wrote:

>> It would be great if we could shift focus and think about the
>> operations impact of depeering vs. just the political and/or
>> contractual ramifications.
>
>
>> Have there been any proposals put forth to the NANOG PC to review
>> this highly visible depeering at the NANOG meeting this month?
>
> Aside from anything else, there is this interesting topic
> on the agenda:
> Abstract: NetFlow-based Traffic Analysis Techniques for Peering  
> Networks
> Richard Steenbergen, nLayer Communications, and Nathan Patrick,  
> Sonic.net
>
> Seems to me that a discussion of traffic analysis could
> handle a slide or two on actual impacts of this depeering.
>
> --Michael Dillon

Here's one way of looking at it:
(copied below b/c the list is not publicly archived)

TV

> From: Tom Vest <tvest at pch.net>
> Date: October 8, 2005 6:00:32 PM EDT
> To: Telecom Regulation & the Internet <CYBERTELECOM- 
> L at LISTSERV.AOL.COM>
> Subject: Re: [CYBERTEL] [ misc fyi ] internet "peering" breaking  
> down (fwd)
>
> Okay now that the flap is over and I have a few minutes to spare,  
> I'll bite.
>
> On Oct 6, 2005, at 10:34 AM, Peter R. wrote:
>
>> Your passionate response deserves a response:
>>
>> It's not very small indeed.
>
> Compared to what?
>
> On 10/1/05, Cogent's network (AS174 -- a very old network)  
> originated the equivalent of  1x /8 + 1x /9 -- that's 1.67% of the  
> "ends" that constitute the global end-to-end network that we call  
> the Internet. Same day/time, Level3's network (AS3356) originated  
> the equivalent 2x /8 + 1x /9 -- or total Internet production 3.05%  
> at that point in time.
>
> Note: numbers are derived from the Route Views archive:
> http://archive.routeviews.org/oix-route-views/2005.10/oix-full- 
> snapshot-2005-10-01-0000.dat.bz2.
>
> In an RFC 1930/2270 compliant world, 99% of networks downstream of  
> either disputant have other, unaffected upstreams, so presumably  
> they don't lose reachability to anyone.
>
> Maybe there are 1b Internet users worldwide, and maybe they are  
> distributed roughly in proportion to the distribution of Internet  
> production. So maybe 5% of the world population was affected by the  
> dispute -- roughly 5m users.
>
> But affected how/how much? If every network end controlled by  
> Cogent and L3 is no more and no less attractive than every other  
> network end, then those 5m users are going to have real problems  
> with roughly 5% of their Internet requirements. In the universe of  
> end-to-end connections, roughly (0.0167 * 0.0205) potential links  
> have been severed --  equivalent to 0.00034235 of the total. If you  
> prefer, make the denominator US Internet production, which is about  
> 60% of the global total on any given recent day. Assume that every  
> US citizen is a user, and cares only about US Internet resources,  
> and you come up with roughly 8% of the national user base having  
> trouble with 8% of their connectivity needs -- that's still one- 
> tenth of one percent of the theoretical (US-US) connectivity total.
>
> And now we know that the problem solved itself in about 48 hours.
>
> Assume that this is excessively simplistic, because of course  
> Cogent and L3 host much more important/active users and content,  
> because there are lots of non-compliant single-homed networks that  
> are also affected (assume also the non-compliant networks are not  
> responsible for their failure to conform to expected use for an  
> ASN). Assume it is unrealistic because some other RFC-compliant  
> networks are multi-homed to tier-2 ISPs that themselves depend  
> significantly on the two parties. Add your own caveats on top of  
> the above; apply your own fudge factor to the numbers until you  
> feel comfortable with them. How skeptical do you have to be, how  
> different do you have to assume the Internet is, or L3 and Cogent  
> are, in order to get to a point where this episode rises to a level  
> of importance sufficient to demand a national or global regulatory  
> solution?
>
>> Many ISPs are single-homed to either one or the other.
>
> What such ISPs contribute to Internet production is either counted  
> among the IP originated directly by AS 174 or AS3356 -- i.e., in  
> the numbers I calculated above -- or they are multi-homed (per RFC  
> 1930), and were not stranded by the depeering -- or else they are  
> out of compliance with the terms under which ASNs are now  
> allocated, under which the broad architecture of the Internet is  
> now interpreted and administered.
>
>> For instance, some Dial-up users of Alleron, purchased by Cogent,  
>> are stranded.
>
> I don't understand what "stranded" could mean in this context.  
> Cogent has many other direct connections to many networks other  
> than Level3. Are you saying that the Cogent-controlled Alleron  
> subscribers had some unique absolute dependency on connectivity  
> with L3, or merely that they, like other Cogent customers, couldn't  
> reach the (3.05%) share of global Internet resources that are  
> directly controlled by L3?
>
> A depeering between *peer* ISPs is not like a phone outage -- even  
> under the most extreme circumstances, you lose only reachability to  
> the Internet resources directly controlled by the counterpart ISP.
>
>> Sites hosted by PSI were stranded, like DrudgeReport. (Poor baby).  
>> It hurts the smaller ISPs more than, say, L3.
>
> Same question, same skepticism. Henceforth, DrudgeReport will  
> consider moving, mirroring, or multi-homing -- and in the mean  
> time, Cogent customers who are regular DrudgeReport readers get a  
> chance to enjoy the remaining 96.95% of the Internet that was  
> completely unaffected, even for them.