Buying IP Bandwidth Across a Peering Exchange
owen at delong.com
Tue Dec 2 23:42:08 UTC 2014
> On Nov 25, 2014, at 10:56 AM, Bill Woodcock <woody at pch.net> wrote:
> On Nov 25, 2014, at 10:47 AM, Colton Conor <colton.conor at gmail.com> wrote:
>> I know typically peering exchanges are made for peering traffic between
>> providers, but can you buy IP transit from a provider on an exchange? An
>> example, buy a 10G port on an exchange, peer 5Gbps of traffic with multiple
>> providers on the exchange, and buy 5Gbps of IP transit from others on the
> Some IXPs have a rule that explicitly disallows this, others encourage it, most don’t care. I don’t know of any that have a mechanism to enforce a rule prohibiting it.
> PCH’s guidance in the IXP formation process is to avoid creating rules which are, practically, unenforceable. So we generally counsel IXPs against having a rule precluding transit across the switch fabric. That said, a crossconnect is a _much better idea_.
>> Some might ask why not get a cross connect to the provider. It is cheaper
>> to buy an port on the exchange (which includes the cross connect to the
>> exchange) than buy multiple cross connects. Plus we are planning on getting
>> a wave to the exchange, and not having any physical routers or switches at
>> the datacenter where the exchange/wave terminates at. Is this possible?
> Yes, it’s possible, but what you describe is a pretty fragile setup. Lots of common points of failure between peering and transit; places where screwing one up would screw both up. If all of this is really tangential to whatever you’re doing, and you don’t mind looking a little out-of-step with best practices, and you don’t mind it all being down at once, any time anything breaks, then it may be a reasonable economy. If you’re planning on actually depending on it, you need to do better engineering, and either spend more money, or allocate your money more conservatively.
> Doing everything the cheapest possible way, regardless of the fragility or complexity, is very short-sighted, and is unlikely to be an economy in the long run.
I’d say that depends…
If it’s an equal cost choice, for example, between getting waves to multiple exchanges and peering with multiple providers at each exchange that way vs. putting a router at one exchange and getting cross-connects there, then I would argue that the former is actually more robust.
More information about the NANOG