Wednesday, July 14, 2010

BSCI - BGP Path Selection Criteria

The following is a list of criteria that BGP uses in order to select 'Best Path'.

Remember that BGP does not perform load balancing. It selects which path is best and then enters that in to the routing table (assuming there isn't an IGP with a better Administrative Distance that has the same route).

The best path is firstly one that is free of loops (AS PATH does not have the same AS listed twice - if it does it must be a loop and the path is discarded). Next there must be a valid next hop, i.e. the router must be able to reach the next hop device. As such there will be a route in the routing table be it connected, a static, dynamically applied via an IGP. Once these 2 points are met then the following factors influence path selection. First match wins and the process stops.

1) Prefer Highest Weight - Cisco proprietary, locally significant, if another router receives the weight flag it is simply ignored.
2) Prefer Highest Local Preference - influences routes in bound to the AS. Use a route-map to set the Local preference (highest wins) then apply it to the nei statement for the routes you which to influence.
3) Prefer routes originated by the local router - these are the routes listed with in the Origin field in #sh ip bgp summary
4) Prefer Shortest AS_Path - least number of hops wins, just like a distance vector routing protocol
5) Prefer lowest Origin Code - where the route originated from be it an IGP > EGP > ? or incomplete which usually indicates the route was received via reoute redistribution from another protocol such as EIGRP.
6) Prefer the Lowest MED - this is effectively the BGP metric and influences how remote AS's view the path of the routes you advertise.

There are many more factors that affect Best Path selection but the BSCI only requires you to know up to point 6 (and I can't be bothered listing the rest as it goes on some what, I think 'Router with tidiest cables in the back of the rack' is the last factor at number 17!? ;o) ) *

*not really

No comments:

Post a Comment