The Logical View

CANS views networks as consisting of applications, services, and data paths connecting the two. CANS extends the notion of a data path, traditionally limited to data transmission between end points, to include application-specific components dynamically injected by end services, applications, or other entities; these components adapt the data path to physical link characteristics of the underlying network and properties of end devices (see Figure 1(a)).

Figure 1: (a) Logical organization of CANS, (b) Physical realization of CANS data paths.

Components are self-contained pieces of code that can perform a particular activity, e.g., protocol conversion or data transcoding. Components operate on typed data streams and are connected with each other based upon compatibility of output and input types (see Section 3 for details). Injected components come in two flavors: stateful services and mobile soft-state objects called drivers. Services extend the original data path to multiple hops, and drivers generalize the traditional notion of a data path to include data transformation in addition to transmission. The primary reason for distinguishing between drivers and services is to ensure efficiency. CANS data paths are created dynamically, using information about user preferences, properties of services and client applications, as well as characteristics of the underlying platform. The components which constitute a data path, the interconnections amongst them, and their internal configuration parameters can all be modified at run time. Modifications are triggered based on either system events (e.g., breaking of a network link) or component-initiated events. The CANS infrastructure provides support to efficiently reconfigure data paths, while preserving application semantics.