Internet Server Execution Models

Models for how to handle incoming Web requests

• fork for each request (NCSA httpd)---expensive!

• pre-fork processes to handle requests---Apache server

• event-driven server. Use a single thread of control in one server to handle all events---use of select(). Avoid context switching between processes. Used in Squid (proxy cache) and others.

• multi-threaded server. Use a single process. Might need a large number of threads. If too many threads then may use a hybrid approach a moderate number of threads each multiplexing a set of connections.

• small set of processes. A main process to handle static content with auxiliary processes to handle generation of dynamic content. Better for fault isolation.