[nsp] ospf default routes and bgp injection

[Steve Lim]

Nick Kraal wrote:
> Steve,
> 
> Thanks for your reply. Redistributing BGP into an IGP and vice-versa are
> known no-nos and an IGP should be used to carry only infrastructure
> networks. Over espresso based arguments with other service providers here,
> was wondering more along the lines of this. For a PoP for example, is there
> any difference running aggregation via something like OSPF summary commands
> in comparison to announcing aggregated prefixes via iBGP.
> 
> Regards,
> 
> -nick/

That boils down to asking "when to use OSPF?" and "when to use BGP?"

	iBGP does not scale well. You have to consider fully meshing your 
network. That's costly, processor intensive and complicated. You could 
implement maybe reflectors or confeds or ATM VCs. But still....

	PoPs with single point of entry/egress should really just be set up as 
stub type areas, announcing into the PoP area only default routes. 
Aggregated routes (ie OSPF summary commands, etc.) of the area can be 
announced into the backbone. OSPF does this quite well, Int IS-IS even 
better. IGP scales well here. iBGP here is prolly just too complicated 
and expensive.

	If you have a multiple-homed PoP, then iBGP has now greater reason to 
be considered. While still more complicated, you now have great 
granularity controlling your metrics. Now, you may have equal cost exit 
points (such as multiple load-shared T1s to aggregate bandwidth.) In 
this case, using an IGP is still preferred, as the metrics are 
identical, and would probably stay identical for the long haul. Though 
EIGRP, which could handle unequal cost metrics. Otherwise, iBGP here is 
probably a better idea, then with an IGP.


Note: NSP community, corrections welcome =)

SL
-- 
<><><><><><><><><><><><><><><><><><><><>
Steve Lim - Network Engineer (Michigan)
Corecomm -An ATX Communications Company
How does a fool and his money get together?