| |
Network Management
For a telecommunications client, we reviewed and recommended ways
to apply artificial intelligence (AI) technologies to network management
and dynamic bandwidth allocation. These AI technologies included case-based reasoning,
neural networks, Bayesian belief networks, fuzzy logic, expert systems,
and planning. SRI is also developing AI technologies for fault management
in large, topologically complex networks. In addition, we are investigating
hybrid technologies that combine different methodologies such as AI, algorithms,
and real-time scheduling.
Network Management for Wireless Communications
Wireless networks
currently provide limited services, mostly voice, but in wireline networks
broadband and real-time services are proliferating. Both commercial and
military applications increasingly need global access to these advanced
services. Thus, protocols are needed to provide them over wireless as
well as wireline networks. Any such protocols must be interoperable with
ATM protocols.
In this project SRI developed new methods for CAC, QoS routing, and multicasting—network
management and/or control functions that are critical for providing robust
communication services with QoS guarantees. Specifically, we developed
(1) an efficient approximation algorithm for the problem of finding the
minimum-cost subgraph that provides two disjoint paths from a source to
each member of a multicast group; (2) the most efficient algorithm that
computes, for a given source and each destination, a pair of maximally
disjoint paths whose minimum bandwidth is maximized; and (3) an efficient
linear programming solution for CAC; this technique learns the best mapping
from the traffic descriptor to the equivalent bandwidth, given a training
set.
Network Modeling and Simulation
The advancement
of computer and communications systems technology presents many challenges
to network planners and managers, such as planning for the expansion of
existing networks, designing new networks, and combining previously separate
networks into hierarchical internetwork architectures. Each of these challenges
requires uncommon engineering expertise and access to appropriate evaluation
tools. Because of the complexity and potential costs involved in network
design and implementation, network professionals are seeking advanced
network modeling and simulation tools to help them make decisions based
on sound analytical techniques.
We have pioneered the use of advanced network modeling and simulation
techniques to solve difficult networking and data communications problems.
As an industry leader in this increasingly vital arena, we have developed
a variety of modeling and simulation tools that help solve both general
network planning problems and problems specific to mobile, military, or
highly dynamic conditions. With these simulation tools, traffic capacity,
the likely effects of different applications upon network loading, and
the location of potential system bottlenecks can be determined before
a network is developed and installed.
Research Internet Gateway
The RIG is a packet
switch that interconnects local-area networks (LANs), via long-haul lines,
to form a national internetwork research testbed. Developed by SRI, the
RIG supports and facilitates experimentation with network control algorithms.
SRI's software implements the Inter-RIG Protocol, which enables researchers
on this testbed to easily substitute new networking algorithms.
|
|
R&D Divisions