Inevitable death, was Re: Verizon Public Policy on Netflix
SNaslund at medline.com
Tue Jul 15 14:33:49 UTC 2014
If you are a multi-homed end user and you feel that a BGP configuration for that is a big management nightmare then you probably should not be running BGP. It would take me somewhere less than 15 minutes to set this up with two carriers and unless the carrier's are at drastically different tiers, there is no need to be doing a ton of "tweaking". I have run a bunch of networks like that and the workload of BGP was not even in my top 100 tasks.
That "awkward and primitive" routing system has scaled pretty well and works well enough that there is not any widespread desire to change it. Sure we might change some things today (which we actually have over time, you know there are different BGP versions, right?), but if you can come up with a better system that is still in widespread use in 30 years, I will be impressed.
Here is the number one reason to have an ASN and your own addresses: If you are using your upstream provider's address space and dump them, you will have to renumber. That is a big deal for anyone with a large internet facing presence and usually results in at least some downtime. Due to the way DNS works (cacheing), there is no really instantaneous way to change all the addressing on your publicly facing systems without incurring some interruption. You also could have your upstream provider get acquired or re-arrange their network whenever they feel necessary and you do not control your own destiny at all. It can also be complex announcing address space you received from one provider through another provider's network especially if those two providers change their peering arrangements between them. As a side benefit of having my own AS number, I can avoid or push traffic to certain carriers by changing my announcements. You can't do that without your own AS.
> An ASN is, literally, just a number. One that's used by a very awkward
> and primitive routing system that requires constant babysitting and
> tweaking and, after lo these many years, still doesn't deliver the
> security or robustness it should. Obtaining this token number (and a
> bunch of IP addresses which is no different, qualitatively, from what
> I already have) would be a large expense that would not produce any
> additional value for my customers but could force me to raise their
> fees -- something which I absolutely do not want to do.
> Perhaps it's best to think of it this way: I'm outsourcing some
> backbone routing functions to my upstreams, which (generously) aren't
> charging me anything extra to do it. In my opinion, that's a good business move.
> As for "peering:" the definition is pretty well established. ISPs do
> it; content providers at the edge do not.
> Netflix is fighting a war of semantics and politics with ISPs. It is
> trying to cling to every least penny it receives and spend none of it
> on the resources it consumes or on making its delivery of content more
> efficient. We have been in conversations with it in which we've asked
> only for it to be equitable and pay us the same amount per customer as
> it pays other ISPs, such as Comcast (since, after all, they should be
> just as valuable to it). It has refused to do even that much. That's
> why talks have, for the moment, broken down and we are looking at other solutions.
> --Brett Glass
More information about the NANOG