The IETF is forming a new working group to address scalability issues in the Internet's routing system caused by companies splitting their network traffic over multiple carriers, a practice called multihoming.
The new working group will build upon a base proposal from a team of Cisco engineers to create a new tunneling mechanism that will be used by the Internet’s edge and core routers.
The new mechanism — dubbed LISP for Locator/Identifier Separation Protocol — is designed to reduce the number of entries in the routing tables stored in the core routers operated by ISPs.
LISP logically separates a block of IP addresses that a company advertises out to the global Internet via its edge routers into two functions: one for identifying the systems using the IP addresses, and the other for locating where these systems connect to the Internet. This separation allows LISP to aggregate the location information, so less of it needs to be stored in the core routers.
LISP works through dynamic encapsulation. Every packet that enters the core routers gets a new IP wrapper that carries information about the destination service provider network, not the end-user IP address. The wrapper is removed from the packet when it gets to the destination service provider.
LISP would operate in conjunction with the Border Gateway Protocol (BGP), which is the primary communications mechanism between edge and core routers.
“The problem we have is that IP addresses are assigned to hosts, and they're not assigned topologically,” says Dino Farinacci, a Cisco Fellow and Senior Software Engineer and one of the authors of the LISP proposal. “This means the core routers on the ISP networks have to carry all of the site-specific routes. We're trying to separate the topological significance of the address from the address allocation procedures…and that will reduce the size of the BGP routing table.”
LISP proponents say the technique also would make it easier for companies to switch carriers without having to acquire new IP addresses because the identification function would remain constant even if the location information changes. And LISP offers companies additional traffic engineering capabilities, backers say.
“More enterprises want to multihome their sites, and they want to do it in a low op-ex way,” Farinacci says. “Today they have to do it with heavy overhead. They have to use BGP, and they have to publish routes into the core. With LISP, we’re putting the routing policy at the edge where the customers can control the bandwidth they pay for.”
Cisco engineer Darrel Lewis, co-chair of the LISP working group, said a key point about LISP is how sites can negotiate their multi-homing policy in an independent, open manner.
LISP developers say the protocol will be deployed as a software upgrade to edge routers, and that no hardware upgrades will be required to run it. They say it will be incrementally deployable and can work with the current version of the Internet Protocol, known as IPv4, or a long-anticipated upgrade known as IPv6.
LISP developers have been working on LISP for two years, and the main LISP document is in its 12th version. Cisco engineers emphasized that they are not making any product announcements about LISP at this time.
“I would characterize LISP as in its adolescence,” says LISP co- author Dave Meyer, a director in the Advanced Research and Technologies Group at Cisco.
LISP developers expect the LISP working group to be chartered by the IETF this summer, after which they will continue to work on prototypes and refine the protocol documents.
The IETF held a preliminary meeting of the LISP working group on Wednesday. The LISP meeting was one of the more well attended sessions.
“If it becomes widely adopted, LISP could be one of the most exciting things to happen at this meeting when we look back on it five years from now,” said long-time IETF participant Paul Hoffman, director of the VPN Consortium.
LISP is not without its detractors. Several attendees at the LISP meeting raised concerns that the approach will increase the complexity of the core routing infrastructure because it adds mapping and tunneling services.
The IETF leadership has made it clear that LISP is an experiment. The draft working group charter says LISP has “potentially harmful side-effects to Internet traffic carried by involved routers” and that LISP is “generally not recommended for deployment beyond experimental situations at this stage.”
Routing table growth
LISP is a byproduct of an effort by the IETF's sister group, the Internet Research Task Force, to re-think the Internet's underlying routing infrastructure.
Both the IRTF's Routing Research Group and LISP have a common goal: to reduce the size of the BGP routing table.
The BGP routing table is a master list of network destinations that is stored in backbone routers and is used to determine the best available path from one network to another. Experts are worried about growth in the BGP routing table, which is straining the processing and memory requirements of the Internet's core routers.
BGP routing table growth is important because it drives up carrier costs by requiring them to keep buying bigger routers with more memory. A key factor driving routing table growth is an increase in companies multihoming their networks to improve reliability.
Geoff Huston, chief scientist at APNIC and an expert on routing table growth, says the BGP routing table has 288,000 entries and is growing at a rate of 14% per year. He estimates that 26,000 enterprises multihome their networks, sending routing table announcements into the Internet's core as they shift network traffic between their ISPs.
So far, Moore's Law — a computer hardware theory that states the number of transistors that can be placed on a chip doubles every year — has allowed routers to handle growth in the routing table entries from multihoming, Huston says. He views LISP as a back-up plan in case routers suddenly can't keep up with routing table growth.
“None of the statistics at the moment point to a looming problem with routing table growth,” Huston says. “Whatever is going to cause routing problems that would require LISP will have to be cataclysmic…If all of a sudden, 300,000 routes becomes 2 million overnight, then the practices we're using now become impossible.”
Today, every core router is aware of every possible destination or routing table entry in case there's a problem. With LISP, core routers only know the routing information needed by the current packet flows.
“The LISP theory sounds good, but in practice it creates a very different network than what we have today,” Huston says, adding that tunneling creates overhead and could slow down network traffic. “LISP changes things around timing and cost. It's much more difficult to understand where and how applications will live.”
Huston's biggest concern about LISP is reliability.
“LISP relies on tunneling, and tunneling is not 100% perfect,” he says. “At times, the tunnel passes a packet that's too big and it disappears without a trace to the sender or the recipient…That's really bad.”
