The following figure shows the system architecture for an ATP environment. The ATP protocol stack consists of four layers:

The Underlying-Network layer

The IBN layer

The ATP layer

The Application layer

Underlying Network Layer

This layer presents the communication infrastructure for our system. This can be an IP infrastructure (such as the case in the Internet), an ad-hoc network (such as the case in MANET), or any other underlying network.

Instance Based Network (IBN) Layer

We define an instance-based network (IBN) as a network of endpoint entities called contents where each content is addressed or located by its name, properties or attributes, independent of its physical location. The content could be a user, an application service, a document, a network node, a network connection or any other object. Unlike IP networks where the IP address is not just a unique ID but also a locator, IBN addressing is decoupled from the location of contents. Contents can actively communicate with each other by sending or receiving messages, or performing a lookup for other contents. Other content types, such as a document, can be passively stored in the network. The IBN layer extends the functionality provided by the current peer-to-peer lookup services (such as CAN, Chord, Pastry, and Tapestry) to achieve its functionality.

Peer-to-peer lookup services provide a mechanism to map a key to some node in network specified by the lookup service, and allows the users to query for these keys. The IBN, however, needs to map a content to a specific node in the network and to route messages to this node. Moreover, the IBN has the unique feature of allowing different instances of the same content to be stored in the network (and hence the name IBN).

more...

Autonomous Transport Protocol (ATP) Layer

The main goal of the Autonomous Transport Protocol is to offer a reliable transport protocol over the IBN. Migration of endpoints should be transparent to each other; that is, the protocol should be able to maintain the connection during the migration of endpoints. The protocol should maintain established connections seamlessly and independent from intermediate node availability. In other words, our proposed protocol should have the same functionality of TCP in static networks while confronting the dynamics of the IBN environment.

more...

Application Layer

ATP-aware applications communicate with the ATP layer to perform migration of the endpoints. The ATP layer is responsible for migrating the open connection state to the new node, while the application is responsible for migrating the application state. To minimize the required changes of the existing applications to be ATP-aware, the interface between the ATP layer and the application layer is similar to the TCP socket interface with the addition of two functions: "migrate" and "land". An application that wants to migrate calls the  migrate  API function to inform the ATP layer of its intention to migrate. When the application migration is done, the application needs to call the  land  API function to notify the ATP layer on the new node to restore the open connections.

Back to Top

[ What is ATP? ] [ System Arch. ] [ References ] [ Members ]

Department of Computer Science 

Last changed: February 20, 2004