Date of Award

Summer 2014

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Director

Ravi Mukkamala

Committee Director

Stephan Olariu

Committee Member

Irwin Levinstein

Committee Member

Michele Weigle

Committee Member

Jin Wang

Abstract

Recently, we have witnessed the emergence of Cloud Computing, a paradigm shift adopted by information technology (IT) companies with a large installed infrastructure base that often goes under-utilized. The unmistakable appeal of cloud computing is that it provides scalable access to computing resources and to a multitude of IT services. Cloud computing and cloud IT services have seen and continue to see a phenomenal adoption rate around the world.

Recently, Professor Olariu and his coworkers through series of research introduced a new concept, Vehicular Cloud Computing. A Vehicular Cloud (VC) is a network of vehicles in a parking lot that can provide computation services to users. In this model each vehicle is a computation node. Some of the applications of a VC include a datacenter at the airport, a data cloud in a parking lot, and a datacenter at the mall.

The defining difference between vehicular and conventional clouds lies in the distributed ownership and, consequently, the unpredictable availability of computational resources. As cars enter and leave the parking lot, new computational resources become available while others depart, creating a dynamic environment where the task of efficiently assigning jobs to cars becomes very challenging.

Our main contribution is a number of scheduling and fault-tolerant job assignment strategies, based on redundancy, that mitigate the effect of resource volatility in vehicular clouds. We offer a theoretical analysis of the expected job completion time in the case where cars do not leave during a checkpoint operation and also in the case where cars may leave while checkpointing is in progress, leading to system failure. A comprehensive set of simulations have shown that our theoretical predictions are accurate.

We considered two different environments for scheduling strategy: deterministic and stochastic. In a deterministic environment the arrival and departure of cars are known. This scenario is for environments like universities where employees should be present at work with known schedules and the university rents out its employees' cars as computation nodes to provide services as a vehicular cloud. We presented a scheduling model for a vehicular cloud based on mixed integer linear programming. This work investigates a job scheduling problem involving non-preemptive tasks with known processing time where job migration is allowed. Assigning a job to resources is valid if the job has been executed fully and continuously (no interruption). A job cannot be executed in parallel. In our approach, the determination of an optimal job schedule can be formulated as maximizing the utilization of VC and minimizing the number of job migrations. Utilization can be calculated as a time period that vehicles have been used as computation resources. For dynamic environment in terms of resource availability, we presented a stochastic model for job assignment. We proposed to make job assignment in VC fault tolerant by using a variant of the checkpointing strategy. Rather than saving the state of the computation, at regular times, the state of the computation is only recorded as needed. Also, since we do not assume a central server that stores checkpointed images, like conventional cloud providers do, in our strategy checkpointing is performed by a car and the resulting image is stored by the car itself. Once the car leaves, the image is lost. We consider two scenarios: in the first one, cars do not leave during checkpointing; in the second one, cars may leave during checkpointing, leading to system failure. Our main contribution is to offer theoretical predictions of the job execution time in both scenarios mentioned above. A comprehensive set of simulations have shown that our theoretical predictions are accurate.

DOI

10.25777/37cf-5q84

ISBN

9781321301892

Share

COinS