Midgard: A Genetic Approach to Adaptive Load Balancing for Distributed Systems
Abstract
The unequal distribution of computational load between the individual nodes in computer networks can be a major source of inefficiency. If communication costs between the nodes on an inter-LAN are sufficiently small, it is possible to achieve gains in overall efficiency by reassigning jobs between nodes in the network to exploit under-utilized computing resources (Stone, 1977). We are currently developing an genetic adaptive approach to this problem. The population in our model is a set of independent but functionally interacting classifier systems which together control the assignment of jobs entering the nodes of an inter-LAN. The population of schedulers must, over time, learn a strategy for migrating the jobs submitted to the network which improves mean network response time. Successive populations, composed of increasingly co-adapted individual schedulers, are produced by the application of abstracted genetic operators. In this paper we discuss our model and its implementation and present some preliminary results.
Cite
Text
Ii and Goodman. "Midgard: A Genetic Approach to Adaptive Load Balancing for Distributed Systems." International Conference on Machine Learning, 1988. doi:10.1016/B978-0-934613-64-4.50023-2Markdown
[Ii and Goodman. "Midgard: A Genetic Approach to Adaptive Load Balancing for Distributed Systems." International Conference on Machine Learning, 1988.](https://mlanthology.org/icml/1988/ii1988icml-midgard/) doi:10.1016/B978-0-934613-64-4.50023-2BibTeX
@inproceedings{ii1988icml-midgard,
title = {{Midgard: A Genetic Approach to Adaptive Load Balancing for Distributed Systems}},
author = {Ii, Adrian V. Sannier and Goodman, Erik D.},
booktitle = {International Conference on Machine Learning},
year = {1988},
pages = {174-180},
doi = {10.1016/B978-0-934613-64-4.50023-2},
url = {https://mlanthology.org/icml/1988/ii1988icml-midgard/}
}