Architectural Styles

Pipes and Filters
We defined this style as an example earlier in this chapter. The Pipes and Filters style is characterized by a single, simple element type (the filter) that processes a data stream, with instances of this type connected by simple connectors known as pipes.

Client/Server
This very widely used style defines a system structure comprised of two types of elements: a server that provides one or more services via a well-defined interface and a client that uses the services as part of its operation. The client and server are typically assumed to reside on different machines in a network (although this is not a requirement of the style—the client and server could be in the same operating system process).

Tiered Computing
A development of the Client/Server style, the Tiered Computing style is widely used in enterprise information systems. A tiered system is considered to contain a number of tiers of computation, which combine to offer a service to an ultimate consumer (e.g., a human user). Each tier acts as a server for its caller and as a client to the next tier in the architecture. A key architectural principle is that a tier can communicate in this way only with the tiers immediately on either side of it; a tier is not aware of the existence of other tiers in the system
apart from its neighbors.

Peer-to-Peer
Often referred to as P2P, this architectural style defines a single type of system element (the peer) and a single type of connector that is a network connection (an interpeer connection). The characteristics of the connector are not important to the style, and the style has been used with a number of types of network connections.

Layered Implementation
The Layered Implementation style identifies a single type of system element: the layer. The style organizes a system’s implementation into a stack of layers, with each layer providing a service to the layer above it and requesting services from the one below it. The layers are ordered by the level of abstraction they represent, with the most abstract (e.g., organization-specific operations) at the top of the stack and the least abstract (e.g., operating system–specific libraries) at the bottom. Depending on the implementation of the style, a layer may be able to communicate directly with any of the layers below it (relaxed layering) or only with the layer directly below it (strict layering).

Publisher/Subscriber
The Publisher/Subscriber style grew out of a realization that client/server interactions are not suitable for all types of distributed system problems. The style defines a single system element (the publisher) that creates information of interest to any number of system elements (the subscribers) that may wish to consume it. A single type of connector, a reliable network link, is used to link the publisher and the subscribers.

Asynchronous Data Replication
While Publisher/Subscriber is normally considered to be a style that allows functional elements to exchange information, a variant of it, Asynchronous Data Replication, is a style used where information in two data stores needs to be kept synchronized (e.g., where it is replicated for performance reasons). The style has three element types: the data source, the data replica, and the replicator. The data source is a data store that owns a particular type of information, while the data replica is a separate data store that wishes to maintain a synchronized copy of some subset of the information in the source. The replicator is the element responsible for recognizing changes or additions to information in the source and performing the synchronization of the replica data store.

Distribution Tree
Another architectural style that relates to data distribution is the Distribution Tree style. This style defines three types of system elements: publishers, distributors, and consumers. The elements of the system are arranged into a tree, connected by network link connectors. A single publisher forms the root of the tree, the distributors are connected to form the intermediate nodes, and the consumers form the leaves of the tree. The publishers publish new or changed information, which the distributors then cache and distribute to their immediate child nodes. If a child node is added or restarted, it can refresh its view of the information from its parent node’s cache. When the information reaches the leaf nodes of the tree, the consumer nodes consume it.

Integration Hub
The Integration Hub style is another data-oriented architectural style, extending Asynchronous Data Replication to situations where information needs to be synchronized between a number of different systems (rather than between replica data stores).

Tuple Space
The Tuple Space style is a type of repository that allows a number of pieces of a distributed system to collaborate to share information. The two types of system elements in the style are the clients (computational elements that create and consume information) and the tuple space itself (a storage area where clients can read and write typed information tuples or records). The clients and the tuple space are connected by a client/server network connection. Clients interact with the tuple space by writing new tuples to it or requesting tuples that match simple search criteria. Clients do not interact directly. Typically the tuple space can also call back to the clients when objects they are interested in change.