North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Why doesn't BGP thread...

  • From: Deepak Jain
  • Date: Sun Nov 10 04:38:09 1996

Maybe I missed something in the original BGP question, but it sounded 
much like this.

A single router has at least two connections to upstream providers [for 
arguement sake, different ones]. BGP is used to negotiate the transfer of 
routing information from their core routers (as opposed to border) into 
this single router. [There can be multiple single routers in this 
description, but the idea is that this single router is a border, and the 
only communications it really passes are internal routing information, 
not external routing information. This router is not a transit router 
between any two upstream providers]

If there are two paths with the same number of hops, across two different 
size/speed/congestion level links, BGP makes a somewhat arbitrary 
decision based on IP address of the interface. 

I don't think this decision is passed on to other routers in the original
question's model. I picture a guy with a T-1 or two, or a 10Mbit FNS
connection and a few T1s. If I am wrong, don't waste your time correcting
me, just ignore this as I am just looking for clarification of the
question; I just want to know whether anything interesting will come out
of this thread or not. 

In this model, this decision is not passed onto other routers [in the 
sense that this single router will in many cases advertise the same set 
of routes to both providers] it is just trying to decide which path to 
send the data down. OSPF [like someone else mentioned] keeps in mind link 
speed issues and automatically adjusts metrics according to some text 
book method. [which can be overriden. I think ether is 1000, and FDDI is 100]
If this router were to look at the BW setting on each interface, and look 
at packet loss on each interface, AOTE, I don't see the problem it would 
cause (either in CPU load, or flapping). Again, in this model, this is 
not a border router at a NAP, this is a router nicely buried in two or 
three bigger networks. I don't think we are dealing with an NP complete 
(Traveling Salesman) type problem either because we are not trying to 
look very "deep" into the upstream network, just 1 link. And again this 
is not being propagated to any other routers.

Any problems with this model?

Of course, if this single router flaps, AND the upstream network 
provider's customer-connection routers do not have half decent access 
lists/filter paths on them, lots of problems could arise, but a lot of 
that is easy to fix.

For example, a network A was receiving network B advertisements from one of 
its customers who also has a connection to network B, network A started 
using the customer's connection for transit to network B because of this. 

Filtering customer route announcements is pretty important for security, 
and flap dampening is essentially just another kind if filtering.

Did I miss any issues?

- - - - - - - - - - - - - - - - -