Expanding Confidence in Networking Simulations. Heidemann, J., Mills, K., & Kumar, S. IEEE Network, 15(5):58--63, September, 2001.
@article{ Heidemann01,
  author = {J. Heidemann and K. Mills and S. Kumar},
  title = {Expanding Confidence in Networking Simulations},
  journal = {IEEE Network},
  year = {2001},
  volume = {15},
  number = {5},
  pages = {58--63},
  month = {September},
  annote = {In May 1999 the National Institute of Standards and Technology (NIST) and the Defense Advanced Research Projects Agency (DARPA) co-sponsored a workshop to discuss approaches to validate network simulations. This article summarizes some of the conclusions of that workshop. As a first step, this article clarifies the difference between verification, validation, and accreditation. It is stated that verification is a process to evaluate how faithfully the implementation of a model matches the developer's intent, as expressed by conceptual descriptions and specifications, provided either in natural language or a formal notation (cf. software functional testing). In contrast to verification, validation relates the model to the real-world phenomenon. More precisely, validation is a process to evaluate how accurately a model reflects the real-world phenomenon that it purports to represent. Accreditation is often used in the absence of technical solutions that can guarantee that a model is free from errors and will provide valid predictions. It denotes a process leading to an official declaration that a given model is fit for its intended use. The successful outcome of most accreditation processes is a written certificate signed by a recognized authority that attests that a prescribed set of processes was correctly applied during the development and testing of a simulation model. As a second step, the article points out some issues to consider when validating network simulations. Very briefly, the article emphasizes the need to clarify what is the baseline; Is it a particular implementation/realization that is the baseline or is the baseline more general? Furthermore, the article points out the importance of accommodating protocol and traffic evolution, as well as the sensitivity of the simulation. As a final step, the article gives some guidelines for successful validation and presents a case study of TCP models. In short, the guidelines states that modelers should compare simulation results with as many alternate representations as possible. This might include laboratory experiments, field tests, as well as analytical models. Furthermore, it is stated that simulation models should be validated in as many contexts as possible and should undergo both statistical and visual validation. It is also stated that simulations must be reproducible and researchers are encouraged to make their simulation models and validation tools publicly available. Finally, in those cases CPU or memory shortage prohibit a full-scale simulation, it is explicitly stated that care must be exercised to avoid introducing artificial boundaries into the model.},
  submitter = {Karl-Johan Grinnemo},
  bibdate = {Monday, May 06, 2002 at 11:31:51 (CEST)}
Downloads: 0