Embedding Games

Large scale distributed computing infrastructures pose challenging resource management problems, which could be addressed by adopting one of two perspectives. On the one hand, the problem could be framed as a global optimization that aims to minimize some notion of system-wide (social) cost. On the other hand, the problem could be framed in a game-theoretic setting whereby rational, selfish users compete for a share of the resources so as to maximize their private utilities with little or no regard for system-wide objectives. This game-theoretic setting is particularly applicable to emerging cloud and grid environments, testbed platforms, and many networking applications. By adopting the first, global optimization perspective, this thesis presents NetEmbed: a framework, associated mechanisms, and implementations that enable the mapping of requested configurations to available infrastructure resources. By adopting the second, game-theoretic perspective, this thesis defines and establishes the premises of two resource acquisition mechanisms: Colocation Games and Trade and Cap. Colocation Games enable the modeling and analysis of the dynamics that result when rational, selfish parties interact in an attempt to minimize the individual costs they incur to secure shared resources necessary to support their application QoS or SLA requirements. Trade and Cap is a market-based scheduling and load-balancing mechanism that facilitates the trading of resources when users have a mixture of rigid and fluid jobs, and incentivizes users to behave in ways that result in better load-balancing of shared resources. In addition to developing their analytical underpinnings, this thesis establishes the viability of NetEmbed, Colocation Games, and Trade and Cap by presenting implementation blueprints and experimental results for many variants of these mechanisms. The results presented in this thesis pave the way for the development of economically-sound resource acquisition and management solutions in two emerging, and increasingly important settings. In pay-as-you-go settings, where pricing is based on usage, this thesis anticipates new service offerings that enable efficient marketplaces in the presence of non-cooperative, selfish agents. In settings where pricing is not a function of usage, this thesis anticipates the development of service offerings that enable trading of usage rights to maximize the utility of a shared infrastructure to its tenants