Gossiptron: Efficient Sharing on the Grid Without Central Coordination

Abstract

To succeed at increasing the value of computational resources by pooling them across participants, grid computing needs a way of distinguishing resource allocations that increase value from ones that do not. I present an algorithm called Gossiptron that is a fully decentralized grid resource allocator and replacement for Oner's peer-to-peer grid scheduler that reduces opportunities for free riding and other abusive patterns of resource use to acceptable levels without getting in the way of effective scheduling. To do so, Gossiptron employs gradual escalation in one-on-one barter relationships, such as in BitTorrent, but extends the concept to transitive barter. I have implemented a simulation of the algorithm, compared results among several of its variants, and analyzed the remaining vulnerability to abuse. Gossiptron maintains effective scheduling and reduces opportunities for abuse to acceptable levels.

Download

---

• The thesis, in PDF (374,196 bytes).

Bibtex Entry

@MastersThesis{,
  author = 	 {Jaap Weel},
  title = 	 {Gossiptron: Efficient Sharing on the Grid Without Central Coordination},
  school = 	 {Vrije Universiteit},
  address = 	 {Amsterdam, The Netherlands},
  year = 	 {2008},
  month = 	 aug
}