The work is motivated by the following observations:
VASSA is a new approach to networking that provides transmission, switching and delivery using multicast techniques that are highly integrated with conventional memory system design, including the use of flat addresses, virtual-to-physical address translation, and cache line units of data transfer.
Communication demultiplexing and resource monitoring are performed in hardware without operating system intervention. Due to this design and the tight coupling with the memory system, VASSA provides efficient, ordered data transfer directly between application address spaces without operating system intervention.
The VASSA connection manager consists of a set of mechanisms to initiate and maintain the state of connections. These mechanisms include flexible routing control, failure recovery, and congestion control.
The connection management mechanism provides a high-degree of source and receiver control over communication services through the control of the route and associated priority. The flexible routing framework integrates datagram network routing and virtual circuit routing into a common framework, as different points along the client/network state spectrum.
Rate-based congestion control protocols are being developed to provide clients a simple means to respond to congestion. Particular emphasis is placed on the accounting-based approach that is supported by the VASSA interface hardware.
In addition, a prototype implementation of the IP+ extension to IP has been developed. IP+ extends IP with support for connection-oriented multicast communication.
An integrated RPC and transport protocol has been developed to exploit the high-flexibility provided by the dissemination-oriented model and the high-performance achieved from the VASSA networking hardware. The design of the libraries based on object-oriented techniques allow applications to specialize protocol behavior based on their unique requirements.
The PARADISE project is developing an architecture for large-scale internetworked simulation environments. New protocols for remote rendering based on position-history-based dead-reckoning and for multicast management based on area-of-interest multicast have been developed. The requirements of the PARADISE project continue to motivate much of the lower-layer research. In addition, PARADISE is being used to evaluate the mechanisms described above.