Task: Describe Distribution
This task defines the deployment architecture for a distributed system in terms of physical nodes and their interconnections.
Purpose
To describe how the functionality of the system is distributed across physical nodes. This task applies only to distributed systems.
Relationships
Main Description

This task defines the deployment architecture for the system in terms of physical nodes and their interconnections. During Task: Architectural Analysis, an initial Deployment Model was defined. In this task, that Deployment Model (specifically the Deployment View) is refined to reflect the current design.

Early in the Elaboration phase, the deployment view is usually quite preliminary, but by late Elaboration it should be well-defined.

Steps
Analyze Distribution Requirements
Purpose To define the extent to which distribution is required for the system. 

Distribution requirements are driven by:

  • Distribution demands in the problem domain (functional requirements) - There may be explicit requirements that the system access or use a specific distributed processor, database, or legacy system to perform part of its functionality.
  • Selected deployment configuration - Specific deployment configurations impose constraints on the system's distribution by defining the number and types of nodes and their interconnections. For example, selection of a multi-tier deployment configuration typically means that you have a client node, a web server node, and an application server node. A specific deployment configuration is usually selected during Task: Architectural Analysis and is then refined during this task.
  • Required resources (nonfunctional requirements) - Time-intensive or computation-intensive functionality might require specific hardware configurations specifically equipped to handle the demands of the functionality; for example, a fast processor, a lot of RAM, or a large amount of disk space. One example of this is digital signal processing, which could require specialized and dedicated processors.
  • The need for fault tolerance (nonfunctional requirements) - The requirement could be to have backup processors.
  • Scalability and flexibility concerns (nonfunctional requirements) - The large numbers of concurrent users are simply too many to support on any single processor. There could be a requirement to load balance the system functionality, thereby providing maximum performance and scalability.
  • Economic concerns - The price performance of smaller, cheaper processors cannot be matched in larger models.

As with many architectural problems, these requirements might be somewhat mutually exclusive. It's not uncommon to have, at least initially, conflicting requirements. Ranking requirements in terms of importance will help resolve the conflict.

Define the Network Configuration
Purpose Define the configuration and topology of the network. 

In this step, the initial Deployment Model (defined in Task: Architectural Analysis) is refined to support the distribution requirements identified in the previous step.

The topology of the network, and the capabilities and characteristics of the processors and devices on the network, will determine the nature and degree of distribution possible in the system.

The following information needs to be captured:

  • the physical layout of the network, including locations
  • the nodes on the network, and their configurations and capabilities (the configuration includes both the hardware and the software installed on the nodes, the number of processors, the amount of disk space, the amount of memory, the amount of swap, and so forth) - hardware installed on the node can be represented using devices
  • the bandwidth of each segment on the network
  • the existence of any redundant pathways on the network (this will aid in providing fault tolerance capabilities)
  • The primary purpose of the node, including:
    • workstation nodes used by users
    • server nodes on what headless processing occurs (to simplify server configuration, server components can be packed into a headless image, which contains no user interface components)
    • special configurations used for development and test
    • other specialized processors
  • IP design and facilities (for example, DNS, VPN), if an IP network exists
  • the part that the Internet plays in the solution

Example

The following diagram illustrates the Deployment View for the ATM

Diagram of the Deployment View for the ATM

Deployment View for the ATM

The diagram illustrates two Nodes (the ATM itself, which is the focus of this example), and the ATM Network Server, through which all connections to the inter-bank network are made. Though the ATM Network Server is out of scope for the builders of the ATM, we show it here to illustrate how network bandwidth can be documented. The diagram also shows the processes and threads which execute on the ATM Node, which are discussed in the next step Allocate system elements to nodes.

Note the use of annotation to document processor and network capacity. Such documentation can also be presented in the documentation fields of the Node (or the devices), in which case it is not displayed in the diagram.

Allocate System Elements to Nodes
Purpose  To distribute the workload of the system. 

In this step, system elements are allocated to the nodes defined in the previous step. Deployment can be described from both a logical and a physical perspective.

Logical deployment is where logical elements (classes, subsystems, or instances of these) are mapped to nodes. These may include threads of control. For example, a logical deployment might state that the AuctionManager subsystem is deployed to the Application server.

Physical deployment is where the files are mapped to nodes. For example, a physical deployment might say that the CloseAuctionTimer.class file is deployed to server76.

Distribution is one area where the sum can be, and usually is, less than the sum of the parts. Achieving real benefits to distribution requires work and careful planning. When deciding which elements will be mapped to which nodes, consider the following:

  • node capacity (in terms of memory and processing power)
  • communication medium bandwidth (bus, LANs, WANs)
  • availability of hardware and communication links, rerouting
  • requirements for redundancy and fault-tolerance
  • response time requirements
  • throughput requirements

Elements are allocated to nodes with the intent of minimizing the amount of cross-network traffic; elements that interact to a great degree should be collocated on the same node; whereas elements that interact less frequently can reside on different nodes. The crucial decision, and one that sometimes requires iteration, is where to draw the line. The distribution of processes across two or more nodes requires a closer examination of the patterns of inter-process communication in the system. Often, there is a naive perception that distribution of processing can off-load work from one machine onto a second. In practice, the additional inter-process communication workload can easily negate any gains made from workload distribution if the process and node boundaries are not considered carefully.

Example

The previous example diagram, the Deployment View for the ATM, illustrates for the ATM Node the allocation of processes onto the node. There is a single process (ATM Main), which in turn consists of three separate threads of control (Customer Interface, ATM Network Interface, and Device Controller).

Some environments provide mechanisms to automate and/or simplify distribution. For example:

  • Clusters: A cluster is a group of servers that act as a unit, typically including functionality such as failover and load balancing. In this case, the Deployment View should describe how system elements are allocated to clusters, as well as how clusters are configured to map to physical nodes.
  • Containers: In component environments, such as J2EE, Microsoft .NET and others, the components execute within a logical computing environment called a container. A container can be considered a "logical node". The deployment view should describe how system elements are deployed to containers, and in turn how containers are allocated to physical nodes.

The use of such supporting distribution mechanisms, and how they need to be configured and mapped to physical nodes to meet the distribution requirements, should be documented as part of the Deployment View.

Properties
Multiple Occurrences
Event-Driven
Ongoing
Optional
Planned
Repeatable
More Information