Friday, January 05, 2007

Directed Diffusion – A New Paradigm for Communication Networks

Traditional networks are based on an IP-style layout. The end points are labeled with unique addresses and a stacked architecture of software provides the communications services sending data from one end of the network to the other. In wired networks this works well but in wireless sensor networks, this paradigm is expensive and difficult to implement.

In directed diffusion, a series of sensor nodes are named with attribute-value pairs. A node requests data by sending an interest request for a piece of named data. For example, a temperature sensor node could have the named value temperature and the attribute of temperature on the sensor. A request for “temperature” would return the numeric value of the sensor. Nodes that do not have temperature sensors but say have pressure or humidity sensors would not respond to the node request for temperature. Through localized interactions, nodes learn how to route information in the most efficient way. Also, repairs can be made by nodes seeking information through broken links such as a node with a run-down battery.

Directed diffusion consists of interests – a query for a specific piece of information, data messages – the requested information, and gradients – the direction state created in each node that receives an interest. Gradients indicate which of the surrounding nodes requested information. Responses to requests are sent back to the requesting nodes.

Interest propagation is the technique whereby the node requesting information sends out a “broadcast” message to the other nodes. The first objective is to see if any of the nodes can make the measurement (say temperature) and if so, then to setup those nodes to start sending their temperature data back to the requesting node. After the requesting node receives data back, it switches from an exploratory mode to a path establishment mode in which it reinforces the path from the requesting node to the nodes providing data. Each node in the chain keeps a data cache that tracks requests and data transmissions. From the data cache it optimizes the path.

One of the advantages of directed diffusion is the ability of the network to “repair” itself by working around failed nodes. When a node sees reduced activity or no-activity it can send information in another direction. This is quite useful in wireless sensor networks in which batteries can run down and EMI interference can render a node inoperable.

As you can see, directed diffusion is a data-centric network. Nodes make requests based on data. Also, nodes communicate only with its neighbors, unlike traditional networks that communicate from one end of the network to the other using routers.

This paper describes in more detail the details of directed diffusion.

Best regards,
Hall T.