A key issue in proxy deployment is delivering client requests to the proxy. The main alternatives include nontransparent deployment, where clients explicitly send requests to the proxy, and transparent deployment, which leaves clients unaware at the IP layer of the proxy's existence.
With nontransparent deployment, the client knows at the IP layer that it speaks to a proxy and not the origin server; therefore, the client must somehow be told to communicate with the proxy. This can be achieved through explicit client configuration, an autoconfiguration file using a configured URL, or autodiscovery of the configuration file. Proxies deployed in a nontransparant manner are called explicit proxies.
Transparent proxy deployment relies on a network element that intercepts requests from clients to origin servers and diverts these requests to the proxy. Proxies deployed using this method are called interception proxies. Request interception can occur at the transport layer (L4) or application layer (L7), and can be accomplished by a switch (L4 or L7 switch), a router, or a special device interposed in a network link (an intercepting link). L4 elements interpret only IP and TCP headers of packets and use destination IP addresses to choose a proxy for the request. Some routers, intercepting links, and L4 switches operate at this layer. L7 elements, represented by L7 switches, can interpret the contents of the request and use this content in choosing a proxy for the request or for implementing fine-grained quality of service policies. However, such content awareness comes at a cost and significantly reduces the throughput of the network element. L4 elements are simpler and leave more processing to the proxy itself but have less sophisticated policies. Recently, intercepting elements that historically operated at the L4 layer have been adding optional content-aware features, while L7 elements have allowed operation in the L4 mode. With this convergence, the choice between L4 and L7 interception is becoming a choice between operation modes rather than products, and the product choice is increasingly based on issues such as price and performance.
Explicit proxies require client configuration and rely on cooperation from the clients. Interception proxies have their own limitations, such as a need for careful placement in the network to avoid a possibility of connection disruptions, disruption of IP-based access control, and more limited reuse of TCP connections. These pitfalls stem from the fact that interception proxies violate the end-to-end principle of the Internet.
For an environment such as an enterprise network where both proxies and clients belong to the same administrative domain, deciding between explicit and interception deployment is a matter of the tradeoff between upholding the end-to-end In-ternet principle versus administrative convenience and proxy enforcement (forcing clients to go through a proxy may be motivated by the desire to monitor or control users' Web surfing). In environments where clients are outside the control of the network administrator deploying the proxy, interception proxies may be the only feasible alternative.