Obtaining an HTML document typically involves several HTTP requests to the Web server, to fetch embedded images, etc. Browsers using HTTP/1.0 [5] send each request on a separate TCP connection. This increases the latency perceived by the client, the number of network packets, and the resource requirements on the server [19,22].
HTTP/1.1 enables browsers to send several HTTP requests to the server on a single TCP connection. In anticipation of receiving further requests, the server keeps the connection open for a configurable interval (typically 15 seconds) after receiving a request . This method amortizes the overhead of establishing a TCP connection (CPU, network packets) over multiple HTTP requests, and it allows for pipelining of requests [19]. Moreover, sending multiple server responses on a single TCP connection in short succession avoids multiple TCP slow-starts [29], thus increasing network utilization and effective bandwidth perceived by the client.
RFC 2068 [11] specifies that for the purpose of backward compatibility, clients and servers using HTTP/1.0 can use persistent connections through an explicit HTTP header. However, for the rest of this paper, HTTP/1.0 connections are assumed not to support persistence. Moreover, this paper does not consider any new features in HTTP/1.1 over HTTP/1.0 other than support for persistent connections and request pipelining.