Enabling Large-scale Network Simulations: A Selective Abstraction Approach

Huang, Polly

Abstract

The Internet research community widely uses network simulators for protocol evaluations. As the Internet's size and complexity double every year, simulation scalability becomes increasingly important. Just like designing a scalable Internet protocol, making a general-purpose network simulator is challenging. Parallel and distributed simulation is one approach to improve simulation scale, but it can require expensize hardware, have high overhead and is limited by the number of CPUs (10-100). In this thesis, we investigate a complimentary solution - abstraction. Just as a custom simulator includes only details necessary for the task at hand, we show how a general simulator can support configurable levels of detail for different simulations, achieving 10-100 times improvements in scale.

Keywords

abstraction
hybrid
centralized computation
end to end packet delivery
finite state automata
survey
FSA

Notes

There are several points of interest in this thesis:
  • It has a thorough survey of most simulation techniques in use
  • A comparison of simulators currently in use
  • It introduces some new techniques from ns
  • Selective abstraction - use of different levels of abstraction in the same simulation
  • A study of how different levels of abstraction affect simulation
  • Finite state automata model of TCP
Huang is mainly concerned with large scale simulations of TCP/IP, and in particular multicast. She is associated with the ns network simulator.
  • Algorithmic routing (have a special case for cycles?)
  • End to end packet delivery
  • Finite State Automata TCP
  • Lili Qiu's session based simulation
A quotation (58): 
In short, how a source releases data into the network will eventually
affect its behaviour in the future. The effect depends heavily on the
specifications of the network. Thus, any model that is generated from 
curve fitting certain traffic measurements will not be valid when applied
to a different network, which is often the case in simulations because of 
limited knowledge about the measured networks.

Related Papers

Bibtex

 @PhdThesis{huang_thesis99,
  author =       "Polly Huang",
  title =        "Enabling Large-scale Network Simulations: A Selective Abstraction Approach",
  school =       "University Of Southern California",
  year =         1999
}
Back to Intro By Author By Importance By Keyword By Title By Reference