Please avoid EGP & BGP backbone peering from the same AS...

Tue Jun 2 14:19:22 UTC 1992

Hi.  Some NSFNET sites are configured to run both EGP and BGP sessions with
the NSFNET from the same AS at the same time.  Please avoid this practice 
and run only one or the other until further notice.  We have discovered 
lost routes and black holing of traffic on the T3 backbone under such a  
configuration. 

An excerpt from Curtis Villamizar <curtis at ans.net>:

 >Subject: mixing BGP and EGP at the same metric
 >
 >The problem arises when there is an EGP route and one or more
 >alternate routes with equal policy metric (either EGP or BGP) and the
 >EGP route is favored.  If the favored EGP route is lost and an
 >alternate BGP route with the same metric exists, the alternate BGP
 >route will always be advertised to the IBGP neighbors.  If an
 >alternate EGP route exists, it will not be advertised and will not be
 >installed.  Depending on which route is found first (hash table
 >position dependent), the alternate route may not be installed.  The
 >more alternate routes at the same metric, the more likely you are to
 >have this problem.
 > 
 >

  --Steve Widmayer
    Merit/NSFNET