Analytical Results on the Performance of Shared Resource Allocation Systems

Licentiate thesis, 2015

Resource allocation is a prevalent problem in a wide range of domains of computer science. Analytical tools that evaluate the performance of resource allocation systems allow us to compare with experimental ones, and utilize the design of such systems. We consider shared-object systems that require their threads to fulfill the system jobs by first acquiring sequentially the objects needed for the jobs and then holding on to them until the job completion. Such systems are in the core of a variety of shared-resource allocation and synchronization systems. We provide methods for estimating the performance of such systems in terms of expected task throughput and delay for completion. To the best of our knowledge, this is a new perspective that can provide better analytical tools for the problem, in order to estimate performance measures similar to ones that can be acquired through experimentation on working systems and simulations. We also study the problem of maximizing the energy utilization in the Smart Grid, where the energy supply becomes available in an online fashion (due to unpredictable energy sources) and the energy demand can have some flexibility (energy dispatch problem). Utilizing a proposed modeling of the energy dispatch problem as an online scheduling problem, we model supply-following demand in terms of the Adwords problem, in order to provide algorithmic solutions of measurable quality. In systems where demands are small compared to the individual supply, we prove a (1-1/e)-competitive ratio. For cases where this does not hold, we extend the Adwords problem to utilize dynamic budgets, and present an algorithm with a 1/2-competitive ratio.