CS 360 - Web Server Architecture

Uniform Resource Locator:
An informal method of identifying a specific location. Used in http, mail, etc. No longer used in technical specifications

Uniform Resource Name:
A persistant method of identifying that doesn't depend on location but instead on resource and organization.

Uniform Resource Identifier:
A generic method of identifying resources

request line: method, URI, version
header lines: additional meta data, method parameters
Optional request body

request line: version, status, phrase
header lines: additional meta data, method parameters Optional response body

Read and parse message
Translate URI to file
Determine if request is authorized
Generate and transmit response
Log request and any errors

(A) To prevent server overloading
(B) To prevent the queue from filling with clients doing nothing.

Web Cache:
Saves pages and content based on different criteria like access frequency and popularity

Anonymizing Requests:
Hides IP address of client
Onion routing: multiple layers of connections encrypted at each step

Filter Content:
Examine URL against blacklist
Examine content against list of banned keywords

Faster load times, lower latency, lower congestion

Reduces server load, lower cost because of less bandwidth required. However, sometimes things shouldn't be cached.

Reduces user load times, also reduces congestion, greater profit for the same price

Reduces congestion and traffic

Check time modified and ETag which is an md5 hash

Web caches at many locations across the country.

It allows the users to be served content much closer to where they are located reducing delay and loss. Amazon

An application that allocates and caches resources. It can help by restricting the critical functions to small, well-tested parts of the code. You can cache files and recent responses reducing load times when accessing them later.

You have a pool of worker threads and one listening thread. New connections are placed into a queue where they wait to be woken up by a worker thread. Must have synchronization between threads.

Spawning and destroying a thread takes time. An excessive number of threads takes memory and context-switching takes time and resources

Threads are more effecient, quicker to switch and quicker to create and destroy.

Only one socket is created. You call epoll_wait to wait for any clients or requests. When you return from epoll wait you process all waiting calls. You don't have to worry about thread syncing because only one process can be inside at a time. Sleeps until event ready

It only fires when an actual event is ready to be processed, so the system doesn't consume resources if there are no events. And it only needs to process the events that actually have been called.

If you were going to have a printer that needed to have multiple clients connected to it. Anytime where you need to have processes that could run at the same time and you don't want to tie up resources with only one path.