[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [RRG] Geographic aggregation-based routing is at odds with reality

On 20 jul 2008, at 16:25, Robin Whittle wrote:


Why would any organisation want to have its choice of address space
constrained by your geo-aggregation idea?
That is a very good question. Our current answer is that the ISPs simply buy larger routers. Would you believe that 12 years ago you could (barely) run BGP with a full table on a Cisco 2500 with a 68020 CPU (at 20 MHz, I believe) and 16 MB RAM? These days I can buy a multilayer switch that can forward packets at gigabit line rate on every port for less than what that 2500 cost, but if I want to run BGP in any meaningful way I have to pay much, much more money.
At some point, the answer is going to be: we don't care about your / 24, we don't have room for it in our tables unless you buy 5 gigabit worth of traffic. At that point the limitations of geo aggregation won't look so bad.
And they're not that bad, really: mainly that when you physically relocate your network, you can't keep your addresses. This is true for most other addresses, too. 


What good does it do them, especially if the constraints change and
require it to change its address assignment when the topology changes?
Addresses don't change when topology changes under normal circumstances. If you are in London with a satellite (or fiber) connection to an ISP in Vancouver, and then you change to an ISP in Milan, you should probably change your addresses. But my assumption is that when we have 10 million multihomers, 99% of those will use ISPs in the same or a neighboring city and this won't be an issue. 


Geo-aggregation is trying to shoehorn the entire addressing and
interconnection arrangements of all providers and end-user networks
into a set of constraints which do not suit those organisations,
No. There is no shoehorning. If aggregation is possible, we aggregate. If aggregation isn't possible, we don't. 


for the sole purpose of enabling the Internet to be connected with
relatively simple routers.



I think this is exactly the wrong approach.
If the shoehorning were necessary, it would still be a good approach: as a general rule, you don't want to optimize for the stuff you need to do once a month or less (changing ISPs) at the expensive of the stuff you have to do every 10 nanoseconds (routing table lookups). 


Organisations need flexibility in choosing their addressing
arrangements,
That statement nicely illustrates our past failures. Organizations shouldn't care about addresses, only about DNS names. 


in choosing who they connect to, in choosing whose
traffic they will accept for transit to another network etc.
No argument there. With geo aggregation this is still possible, although of course some choices won't give you good levels of aggregation. A new option would be that you can easily buy or traffic engineer connectivity towards different regions from different ISPs because regions are now identified by aggregates. 


Geo-aggregation forces humanity to bend to meet the needs of a
dumbed down routing system.
Hm, that's something I should think about on my flight to Dublin. </ irony> 


It is non-trivial thinking of a better routing and addressing
system, but some folks have tried: LISP, APT, Ivip, TRRP, Six-One
Router.


Shim6?  :-)


Map encap schemes have faced criticism from some folks because they
would be hard to scale to billions of EIDs.
Well, that's just a challenge, if we do our work we should be able to solve that.
The real criticism is that we ALREADY have end-user names and infrastructure addresses: FQDNs and IP addresses. Apparently that wasn't enough. How do we know that a third naming layer will do the trick? 

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg