HTTP Persistent Connections

Copied from http://www.jmarshall.com/easy/http/#postmethod

In HTTP 1.0 and before, TCP connections are closed after each request and response, so each resource to be retrieved requires its own connection. Opening and closing TCP connections takes a substantial amount of CPU time, bandwidth, and memory. In practice, most Web pages consist of several files on the same server, so much can be saved by allowing several requests and responses to be sent through a single persistent connection.

Persistent connections are the default in HTTP 1.1, so nothing special is required to use them. Just open a connection and send several requests in series (called pipelining), and read the responses in the same order as the requests were sent. If you do this, be very careful to read the correct length of each response, to separate them correctly.

If a client includes the "Connection: close" header in the request, then the connection will be closed after the corresponding response. Use this if you don't support persistent connections, or if you know a request will be the last on its connection. Similarly, if a response contains this header, then the server will close the connection following that response, and the client shouldn't send any more requests through that connection.

A server might close the connection before all responses are sent, so a client must keep track of requests and resend them as needed. When resending, don't pipeline the requests until you know the connection is persistent. Don't pipeline at all if you know the server won't support persistent connections (like if it uses HTTP 1.0, based on a previous response).