May 7, 2006

Before you begin your journey into simulation, consider the followings.

To follow up on this:

Make a simulation plan first. What do I mean by that? I mean that, before you implement anything, think about WHAT you want to test and how you could show that. I am currently attending a lecture on simulation and our professor really stressed that most people doing computer simulations focus too much on the implementation and forget about the planning. He also said that planning should be 30-40% of the total work...

If you are into wireless routing protocol simulation, these are the things you should consider :

- ask myself what are the strengths/weaknesses of these protocols. For this you'll have to read the theorethical papers describing them and to understand how they are optimizing the network. Basically, routing is an optimization problem but it is not defined what to optimize for. eg.

  • quick set-up of routes?
  • guaranteeing route correctness?
  • little signalling overhead?
  • quality of routes once they are set up? etc.

    - now compare the theorethical goals of these protocols with each other.
    How could they be compared in a live setting? eg. is one of them tuned to a specific situation and the other one more to another scenario? Can they be compared at all or would you be comparing apples with oranges?

    How can you set up "obstacles" or bottlenecks for the protocols to see if they deal effectively with these difficulties?

    - definitely check the papers "MANET simulations - "the incredibles" by Kurkowski, Camp, and Colagrosso as well as "The mistaken axioms of wireless-network research" by Kotz, Newport, and Elliot.

    Think very hard about the PHY layer model you are going to use and whether/how this PHY model will influence your results.

    If you check the mailing list, you will find that some newer implementations have been announced - maybe you'll need to check them out.

    - to make your own life more simple: how can you seperate concerns so you really can test for cause and effect. I made this mistake in my own simulations; I tried to set up a "typical" scenario which got very, very complicated and I couldn't show anything anymore because the effects I saw were caused by multiple causes which all interacted/interfered with each other.

    I have now gone one step back and try to work out simple scenarios which test for only one effect first. When I have a better
    understanding (after doing several simple tests) I will try the complex scenario again. With the insight gained through the simple settings I might even be able to interpret what is happening there then :-)

    - only AFTER you have thought theoretically about all that and planned your simulation, sit down and actually start coding.

    Tags :

  • No comments: