Representationl State Transfer (REST) is a term to describe an architectural style of sharing information with consumers using already existing protocols such as HTTP. In the strictest sense of the term, the transport protocol need not be HTTP.
REST was first coined by Roy Fielding in his
year 2000 doctoral thesis. Unlike things like Simple Object Access Protocol (SOAP), Distributed Component Object Model (DCOM), Common Object Request Broker Architecture (CORBA), it is not an architecture nor a protocol but rather a style
of architecture. As a result, a lot of things that don't call themselves RESTFUL full under that umbrella or use some of the same concepts, or can arguably fall under that umbrella.
What is probably so alluring about REST is that it gives a catchy name to something a lot of people were doing already and describes how much of the web operates.
Contrary to some popular belief, it is not NEW technology but rather a grouping of mainstream technology with a flashier name. Part of this confusion is that the cult following of REST is a fairly new phenomenon although the underpinnings are relatively old. The REST movement reflects a return back to the basics that prioritizes simplicity and accessibility over complexity and formality.
Nowadays REST is most often used to refer to web services where resources are requested via plain URIs and GET requests, representations are returned in simple XML or JSON format and resources are created using POST, updated using PUT and deleted using DELETE HTTP verbs. This is similar to XML-RPC except that
XML-RPC has a concept of state and everything is generally encoded in an XML message envelop. XML-RPC also uses POST for both updating and accessing resources unlike REST which tends to use GETS and URIS for resource access. The advantage of using get is that resources can be bookmarked. SOAP is similar to XML-RPC and in fact was born from the XML-RPC standard except the XML message streams are more complicated and formal, but arguably richer in functionality.
Contrary to some popular belief, REST is not a silver bullet nor was it designed to be. It doesn't work for all problems and web applications. Our personal opinion: REST is well suited for transporting data that will be consumed by various kinds
of clients, but is not well suited for updating of data or where authenticated transactions are needed.
REST has 4 basic features that differentiate/and it shares with similar Architectural styles.
Client Stateless Server - As the name suggests - the state of an object is part of the message, and is commonly referred to as a stateless communication. It is not done with things like Session cookies where the
server maintains some stateful view of the client and the client passes a session cookie saying (here is my ticket - give me my state). The server does not hold information about state, only the client.
Right away one can tell - this can not work for all modes of communication that require immense amounts of state information to be maintained, but does have the advantage of should the application server hiccup or connection to the server times out or breaks only the current message is lost.
It also works well for Web-Farms that are simply outputing data since the need for such Web-Farms to replicate state is not needed (think image caching networks such as Akamai).
Client-Cache - The idea of client caching. The server can dictate certain requests as being cacheable and if cacheable a client can use the cache request to satisfy future similar requests instead of going back to the server. This saves on band-width but has disadvantage of possibly resulting in stale results. Keep in mind again this concept is not
new and most webservers are designed to work that way and pass this info via http headers.
Layered System - two way interaction. In a REST style architecture, there is a client and a server. The client is only dependent on the server it communicates with. It has no knowledge of the components the server uses to fulfill its request.
That server can be a client in another REST interaction and keep its own cache to serve up like requests. This does not break the client keeps the cache rule as the server is acting as a client in this context. Think DNS. DNS is a perfect example of such a style where intermediaries cache
requests for a certain period of time and act as clients to DNS servers further up the root and behave as servers to DNS and client computers below.
Resource and Resource Identifiers - REST is predominantly a mechanism for accessing resources although it can be used for editing as well. The
key element of it is a mechanism for defining resources,
how a resource or grouping of resources are requested via a Resource Identifier (URL or URN), transfer of representation via (HTML, XML, Jpeg etc.),
representational metadata (e.g. media type, last modified), control data (such as how long it can be cached). Yes this is pretty much
a common concept in web interfaces.