Scalable Web Request Routing with MPLS

High-volume Web server and cache installations achieve scalability and reliability by using a front-end dispatcher to route incoming client requests among a cluster of server machines. Dispatchers typically operate at layer-4, using transport-layer information (e.g., IP address/port), or at layer-7 using application-layer information (e.g., HTTP headers), to direct clients to the appropriate server. Layer-7 dispatchers, while more flexible than layer-4 approaches, suffer from limitations on scalability and performance since they must perform TCP connection termination and management for a large number of clients. In this article we describe an approach that, when combined with an intelligent client-side proxy, can implement a dispatcher using commercial, high-performance, off-the-shelf switching hardware, while also providing the flexibility of a content-aware router. We leverage the growing migration of networks to Multiprotocol Label Switching (MPLS) in order to provide
more flexible routing, but rather than using labels to express routing and forwarding policies, our scheme maps application-layer information onto labels to enable high-performance Web request routing. Unlike con-ventional schemes, our technique assigns some of the dispatch function to an MPLS-aware client-side proxy that applies the appropriate label for a given connection. This approach removes the main bottleneck from the system (i.e. TCP connection termination at the dispatcher), and lends itself to realization in a standard MPLS switch, thus obviating the need for costly, specialized layer-7 Web switch hardware while providing the same functions at a much improved price-to-performance ratio.