[nycbug-talk] Cogent and Sprint - a signal of things getting Oldschool?

[Isaac Levy]

On Nov 1, 2008, at 6:29 PM, Marc Spitzer wrote:
>>
>> This thread, and Miles' earlier email about 'hot potato' routing,
>> makes me ask perhaps a stupid question:
>>
>> Why is routing not synchronous?  Why is sending more expensive than
>> receiving packets- from a transit perspective?
>
> Could you expand on what you mean by synchronous?
>
> marc

Alex corrected me below, symmetric, and answered my question,

On Nov 1, 2008, at 6:33 PM, Alex Pilosov wrote:

> On Sat, 1 Nov 2008, Isaac Levy wrote:
>
>> This thread, and Miles' earlier email about 'hot potato' routing,
>> makes me ask perhaps a stupid question:
>>
>> Why is routing not synchronous?  Why is sending more expensive than
>> receiving packets- from a transit perspective?
> You meant 'why is it not symmetric'. It's not supposed to be. Just  
> look
> how BGP works - each AS "announces" a set of networks, and  
> determines from
> a set of available AS-PATHs (based on some policy, involving "money"  
> or
> as-path-distance), where the outgoing packet will go. If a packet is  
> going
> from AS A to AS B, the set of AS-PATHs and associated policies will be
> very different and unlikely to be symmetric vs packets going from AS  
> A to
> AS B.
>
> There's no such thing as 'sending vs receiving' - in any TCP
> conversations, packets always flow both ways. What you mean, and  
> what is
> being sort-of-used-as-justification-of-higher-expense-for-eyeball- 
> network
> is the fact that
>
> a) Almost everyone uses "hot potato" routing - that means, you get  
> it out
> of your network as soon as you can to pass it off to peer/transit/etc.
> Assuming you have multiple points of interconnection with it, for  
> example,
> if Pilosoft had nationwide network and connection to AT&T in San  
> Jose and
> NY, and I have a packet from NYC customer going to AT&T, I *should*  
> still
> hand it to AT&T in NY, regardless of where it is going to end up. [1]
>
> In fact, hot potato is the *preferred* way - since you don't know  
> how your
> peer/transit network works, you should optimize *your* path, and let
> *them* worry how to carry it inside their network.
>
> b) As result, assuming AT&T customer is downloading something hosted  
> on my
> network, AT&T will carry large packets with content on the path from
> NYC->SJC, and I would carry ACKs SJC->NYC. Since ACKs are much  
> smaller,
> the load on their network will be more than mine.
>
> [1] There's such a thing as "MED" (multiple exit discriminators) to  
> avoid
> "hot potato" routing, but it has potential to fuck things up more -  
> so it
> is rarely used except in situations where your peers *require* you  
> to obey
> the MEDs


Wow- I get it.  Excellent explanation.

Alex: I guess it seems I don't understand BGP use as much as I  
should.  Can you point me to any urls/books/whatever about BGP that I  
could check out?

Rocket-
.ike