Grid Computing Research Laboratory State University of New York (SUNY) Binghamton Department of Computer Science

Deger Cenk Erdil and Michael J. Lewis
"Supporting Self-Organization in Hybrid Grids",
Self-Organization in Pervasive Distributed Systems, 23rd Annual Symposium on Applied Computing (SAC08),
Fortaleza, Ceara, Brazil, March 16-20, 2008.
[PDF] [bibtex]

Abstract
Increasing scale, dynamism, and complexity of hybrid grids make traditional grid resource scheduling approaches difficult. In such grids, where resource volatility and dynamism is common, self-organization is a key technique for autonomous grid nodes to follow basic rules to minimize human participation, and administrative bottlenecks. This paper presents experimental results with a framework for distributed grid resource scheduling. In particular, we study information dissemination, which distributes information about dynamic grid resource states to remote schedulers. The framework helps each autonomous grid node self-organize by (1) self-configuring its operational parameters based on dynamic grid characteristics, and (2) self-adjusting its dissemination behavior by taking feedback from the system. The framework also helps distributed grid schedulers to find the tradeoff between two important performance parameters: dissemination overhead, and query satisfaction rates. We show by simulation that autonomous grid nodes that self-organize into small groups, and compare their local state to the states of other peer nodes, can perform comparable to both (1) similar dissemination protocols that are statically configured for each specific case, and (2) a theoretical central metascheduler that operates on complete knowledge of available resource and offered load states.