Transcript Protocols and Systems for Agile, Interactive and Intelligent Routers

High Rate Data Delivery Program
September 2002
Intelligent and Agile Protocols and Architectures for
Space and Terrestrial Networks
Funda Ergun
Behnam Malakooti
Case Western Reserve University
NAG3-2578
Objective
• Given a space, terrestrial, or hybrid network consisting of
aerospace, near earth and terrestrial components, design
architectures and protocols which will allow for faster, more
reliable routing with Quality of Service, learning from their
experience.
Nature of the Problem and the Solutions
• Network composed of various types of components—satellites,
terrestrial routers, end hosts, mobile devices, …
• Network topology may change in time
• Different transmissions require different parameters
• Existing protocols provide basic service, not much more
Examples of Existing Service
• IP, routing using BGP. Routing performed using shortest path or
policy routing. Topology discovery is crude and takes time.
• Label switching, MPLS. Path allocation, backup paths are
supported. Main concern is reliability, main goal is to keep
packet header short.
Desired Routing Algorithm
We would like our routing algorithm to:
• Respect bandwidth, reliability, delay, cost concerns.
• Be adaptable to changes in topology.
• Not require much computational power from intermediate nodes.
• Incorporate policy routing if necessary.
Design of a routing protocol
•
•
Routing/scheduling algorithm with desired properties is
intractable.
One can resort to heuristics/approximation algorithms.
Issues to be dealt with:
Running time with realistic data
How “good” the routing is performed.
Techniques Used
• Rounding and scaling of data for approximation
• Lagrangian relaxation
• Estimating future resource allocation.
Simulation results are very promising for both multicast and unicast
networks: algorithms/heuristics are extremely fast, require very
little computation, and route in a very “intelligent and precise”
way.
Simulation Results
• If all of desired values stay within bounds, our algorithms find
solutions that cost up to 5% more than the optimal solution in
around 100msec for 50-100 node networks for unicast, up to
15% more for multicast networks.
• Finding the optimal multicast routing for a 10-node network
takes a full day; at 12 nodes it cannot be done.
[ESZ], [ESZ2], [WEX].
Protocols
• Message passing and data storage requirements: most of the
data is kept in the routers.
• Not all nodes need to have access to all the information. Using
hierarchical routing, data dissemination needs can be
minimized.
• Header lengths need not become very long; MPLS-like
structure.
• Routing data can be translated into diffserv-like class system.
Intelligent Mobile Decision Makers
•Developing a framework for Intelligent Mobile Decision
Makers as a vehicle to achieve distributed decision
making.
•Developing Intelligent Mobile Decision Makers that can
operate within IP.
•Developing an approach for Clustering of Packets: for
more effective and flexible network communication.