Jakub Gąsior, PhD

The objective of this project was to address the problem of efficient resource management in Cloud Computing (CC) systems, aiming to allocate and schedule client's submission to available machines in such a way that providers achieve optimal resource utilization while users meet their applications’ performance and security requirements with a minimum expenditure and overhead. To this aim we developed a number of intelligent scheduling and load balancing mechanisms for the optimal spread of workload on distributed machines and servers of the considered Cloud platform taking into account two general criteria: the execution time of the submitted tasks and the level of provided security assurance. 

We reformulated classic assumptions concerning load balancing and scheduling algorithms in the following way: (a) we assumed that available computational resources are heterogeneous and the computational power of resources depends on a processor's speed; (b) clients realize their own goals and generate computational workload characterized by a parallel job model, while computational load and availability of resources may change dynamically in an unpredictable manner; (c) computational nodes of the CC platform are characterized by a quantifiable level of reliability; and finally, (d) there exist multiple levels of coordination and oversight over the whole CC system.

Based on the above-mentioned assumptions, we developed a mathematical model of the Cloud facilitating the analysis of the performance of the whole system as well as the verification of robustness against faults, efficiency, scalability and interoperability. We worked out solutions for security-aware computation in CC systems with an application of so called "soft security" mechanisms based on metaheuristic approaches employing Game Theory, Cellular Automata models and nature-based Multiobjective Optimization. We studied three concrete models based on this general framework exploring a different facet of the rules that govern how various entities might engage in collaboration in practice. We also showed how to obtain allocation schedules with good trade-offs between the results achieved by the participants under three sets of rules, varying in the amount of information necessary to undertake a scheduling decision. 

Proposed models have been implemented and thoroughly tested experimentally. A comparison with other well known scheduling and load balancing algorithms has confirmed the advantages of the proposed solution. The details of the proposed allocation schemes have been described in a number of papers (cf., here). 

You are here: Home Results Jakub Gąsior, PhD