Packet Design Intros BST
A well documented Internet problem relates to Border Gateway Protocol (BGP) routing tables, the directories of where things are located on the Internet that enable big backbone routers to do their job (see Experts Sound Alarm on Internet Routing ). The Internet's growth -- and the millions of new devices with IP addresses -- has caused these BGP tables to grow to about the size of Anna Nicole Smith. Also, more enterprises and ISPs are using BGP to connect to multiple ISPs, which adds significantly to the number of BGP route table entries.
Though today's routers are extremely fast, a lot of their processor power and memory are consumed when the BGP-related information gets too big. "What's essentially happening with BGP is these massive address tables -- phone books, if you will -- are being moved from one ISP to another to another to another," says Richard Clarke, special advisor to the President for cyberspace security.
Packet Design's Estrin says BGP's problems can be addressed by changing the way BGP information is carried to the other routers in a network. Right now BGP tables are carried via TCP (Transmission Control Protocol), a protocol that makes a point-to-point connection between routers.
Enter BST, or BGP Scalable Transport, which Packet Design describes as an alternative transport protocol for BGP. TCP devotes memory to keeping track of the state of the TCP connection. But rather than keep a plethora of point-to-point links, BST works by using a technique known as "flooding" to send BGP tables only to a router's immediate neighbor routers. Those routers, in turn, send it to their neighbors, and so on. Theoretically, fewer resources are exhausted this way than keeping track of the state of so many TCP connections within a network.
BST comes as a reference source code module. The reference product contains all of the tools a router vendor would need to build and test BST in a FreeBSD (Unix) environment. BST doesn't change the BGP protocol, Packet Design says. But a router's BGP implementation would need some tinkering, so the router will use BST for message passing instead of TCP.
Estrin says there are also security benefits to this approach as well. In a BST-enabled cloud of routers, only one router at a time would have its IP address exposed to the outside world. When that one router fails, other routers would step up, one at a time, as designated by a network administrator, to take its place.
"There's a lot of focus in the Internet on making things faster," Estrin says. "What we're finding now is that we need to focus on the control plane, which allows you to make it better, not just faster."
Estrin says Packet Design aims to sell BST to router vendors and pricing starts at $100,000. This is the company's second commercial product since inception. The first was a router network troubleshooting system called Route Explorer (see Packet Design's Routing 'Spy' ).
Though the potential customer base is small, and it may be a tough sell, vendors might take solace in knowing that BST's creator is former Cisco chief scientist Van Jacobson. "As a person who has played a large role in the development of TCP over the years [Jacobson] is a credible person to address the problems of TCP," says Mark Seery, an analyst at RHK Inc.
Whether it’s a commercial success or not, the fact that someone is attempting to solve some of the problems surrounding BGP is getting some cheers. "Even if all [Packet Design] does is open up a conversation on the topic, I think [it] will have done the industry an incredible service," Seery says.
Other routing experts say BST is frivolous business. "BGP tables are big and TCP has high memory costs, but so what? Memory is cheap, and even newer edge routers are scaling to hold tables many times the size of today's full Internet table," says one routing wonk, who asked not to be named. "Also, the rate of BGP table growth is slowing." — Phil Harvey, Senior Editor, Light Reading