Using BGP as an IGP at all Layers of the Network

[rich_foster]

I am in a ferocious debate with someone who insists on using BGP across their entire network - and is a global standard.  This includes eBGP sessions between the core and aggregation layer in the DC, the core and user distribution layer, and between everything else.  Every layer / block / silo is in a different private AS with eBGP peering, even wiring closets are eBGP connected.  Confederations are not used and eBGP has basically replaced OSPF.

I have searched every Cisco design document, Cisco Press publication and forum I can find and have not even found a mention of anything even near.  The best argument I’ve heard is the BGP route manipulation features which I would only consider valid when connecting large sections of a network, internet peering, MPLS VPN peering, etc.

I have no doubt that this is a terrible design I’d just like to know if anyone has seen anything similar anywhere, ever?!  Can anyone think of one single positive reason for doing this?

Any responses much appreciated - full points for any responses!

[Reza Sharifi]

Rich,

As you noted, BGP is usually used in service provide or campus environment where there is a need to run MPLS L2 or L3 VPNs for separating customer traffic. If you just have simple design and no need to run VRFs, then I don't see a need to run BGP.  When you run BGP, you still need an IGP, i.e. ISIS, OSPF, RIP,etc...It is correct that BGP gives to capability of route manipulation, but than again that is mostly used in large service provider environment where they have 100s of peering with other providers or customers.

HTH

Reza

[milan.kulik]

Hi,

IMHO, this question was discussed in this forum in the past, try to search.

There are reasons why BGP is NOT recommended to replace an IGP, the most evident are:

1) the full-mesh necessary for all iBGP-connected routers if running without any IGP

2) slow convergence when default timers left

3) not  available on low-end routers or L3 switches

4) configuration complexity compared with IGP protocols

5) suboptimal routing path chosen often compared to the results of an IGP (considering line bandwidth, e.g.)

BR,

Milan

[Giuseppe Larosa]

Hello Richard,

I have seen once a network made in this way with private AS numbers and no IGP at all and even the use of multiple eBGP or iBGP sessions within each network portion.

The only advantage was that they were able to deploy for example NAMs only on the active routing plane because with default settings only the best path is used ( without tuning with maximum paths).

However, routing convergence was slow and core network had 600 aggregate networks in the BGP table and whatever failure happened you couldn't see any prefix to be withdrawn.

That customer was a bank and it could afford to use one of 4 links between network blocks.

Deploying an IGP for infrastructure and to use BGP for advertising services may help to have a network MPLS ready, for an enterprise an IGP can be enough.

Hope to help

Giuseppe

[rich_foster]

Giuseppe, completely agree regarding MPLS, but if you were implementing BGP to support MPLS VPNs you’d presumably only want to use iBGP and not have loads of AS’s?

Thanks