Why use BGP instead of OSPF/EIGRP on the WAN?

Why do we use BGP as the preffered EGP? What makes it better then OSPF/EIGRP? Why use BGP between you and the provider, to exchange routes?

Find more posts tagged with

Comments

networker050184

Stability, scalability and ease of policy.

FrankGuthrie

networker050184 wrote: »

Stability, scalability and ease of policy.

OSPF is not stable? Als what do you mean with the term scalability?

Ease of policy, what doe you mean with this?

networker050184

OSPF is stable to a certain point, but it's designed to be quick. The more prefixes you put into a routing protocol that needs to be quick the more load you put on the device in a failure scenario. Imagine OSPF flapping 10 times in a second while the router has to continually insert and withdraw a hundred thousand routes. Not good. BGP can scale for pretty much the same reasons.

Ease of policy means filtering of routes, preferring routes over others, etc. BGP has many, many knobs to turn to get traffic how you want it. OSPF on the other hand has many rules to jump through when filtering. You also have a very small amount of knobs to turn to when needing granular preference of routes.

There are instances when OSPF/EIGRP is used on the WAN. Usually when you have a small number of prefixes. Most providers prefer to run BGP for the reasons above though.

gorebrush

I would add granularity to the list as well. One can exert precise levels of control over BGP.

FrankGuthrie

gorebrush wrote: »

I would add granularity to the list as well. One can exert precise levels of control over BGP.

In what way can you exert more control. Isn't OSPF also tweakable, same goes for EIGRP?

Simrid

BGP has many attributes, such as weight and local preference. This enables routes to be manipulated a lot more easily as you can implement route maps and use it along side some policy based routing.

Whilst OPSF and EIGRP does let you tweak some attributes, they are best suited for the LAN, as mentioned by OP, OSPF needs to billd a logical topology tree and LSA's get broadcast on a per area basis.

dppagc

Furthermore BGP separates different AS. So the other AS will be under the control of another ISP.
Those routes will not be in your routing table and not your business.

daveyb

dppagc wrote: »

Furthermore BGP separates different AS. So the other AS will be under the control of another ISP.
Those routes will not be in your routing table and not your business.

Routes from another AS will most likely be in your routing table... if they are not, you probably don't have full reachability...

The main reason to use BGP is how it scales. A common service provider design is to have only loopbacks and point-to-point addresses in your IGP, and everything else in BGP. This is because under failure conditions, your IGP can converge quickly, and provide the next hop routes that BGP requires to route traffic, minimising disruption. Coupled with features like LFA you can really get convergence sorted pretty quick.

To run an IGP with your provider, you would need some kind of globally connected area... who would be area 0 (or level 2 in IS-IS)? If a link flapped in Australia, everyone's routers would reconverge. Today's routing table is around 600k prefixes. Imagine the length of time it would take for OSPF/IS-IS (Seriously, no one seriously uses EIGRP in production at scale...) to converge under a link failure. Yikes!

fredrikjj

daveyb wrote: »

To run an IGP with your provider, you would need some kind of globally connected area... who would be area 0 (or level 2 in IS-IS)? If a link flapped in Australia, everyone's routers would reconverge. Today's routing table is around 600k prefixes. Imagine the length of time it would take for OSPF/IS-IS (Seriously, no one seriously uses EIGRP in production at scale...) to converge under a link failure. Yikes!

No, you would use redistribution, and the network would never converge, just like BGP doesn't converge globally; there are always changes. According to Gredler's ISIS book (The Complete Is-Is Routing Protocol by Hannes Gredler), IS-IS was originally intended to carry external information as well, but like people have mentioned already in this thread ISIS/OSPF don't scale to that point, and you would lose BGP's policy features. There are too many prefixes now, and the link state protocols are focused on transparent flooding without flow control, which will break once you have enough data to transfer. That BGP uses TCP makes it more robust in that regard. But, if the global table was say 10,000 prefixes, it would be technically possibly to use only ISIS/OSPF.