Configuration, operation, reliability, scalability and security problems with BGP have all grown as the Internet has grown. Based on detailed analysis of EBGP data from RIPE RIS and IBGP data from collaborations with several tier one ISPs, we believe that many of the most serious problems with BGP are not in the BGP protocol but are instead due to BGP's use of point-to-point TCP connections for its transport. Transport problems are much easier to fix than protocol problems since a new, parallel transport stack can be added to existing BGP implementations in an incrementally deployable, evolutionary fashion without having any impact on configuration or route selection behavior.
Using the insights gained from our IGP and BGP measurement studies (reported at NANOG and elsewhere) and our experience developing different styles of transport protocols (such as SRM and ALF) we set about designing and implementing a new BGP transport. We call the result BST (BGP Scalable Transport). We will describe the implementation and discuss the issues underlying the most significant aspects of the design. We will show how it solves many of the existing problems with BGP and how, with a few small additions, it can form the basis of new BGP implementations that are extremely easy to configure yet highly reliable, scalable and secure.