Impatient job scheduling under cloud computing

Cloud computing, a new concept, refers to a hosted computational environment that can provide elastic computation and storage services for users per demand. This paradigm arises with the huge growth in applications and data sizes. Many agencies, organizations and departments are responsible for time critical tasks and these tasks need to be completed as soon as possible. At the same time, these agencies also face IT problems because of the huge growth of applications, data and solution sizes. Many experts proposed that cloud computing is a solution to these problems such that each agency can execute its tasks via the cloud and expand their requirements based on the situation. In this thesis, a study on the scheduling of impatient jobs in a cloud environment is presented. The study can be divided into three parts. The first part focuses on review- ing the previous immediate mode scheduling and adopting them on cloud paradigm. The limitations of those algorithms were addressed and this leads to the proposition of an algorithm that has the ability to map the impatient jobs to virtual machines near its input, output, application, or forth party. The challenge was how to consider the file sharing in on-the-fly way. This algorithm has been improved to take into account the input file sharing after analyzing the effect of this case on system performance. The proposed algorithm is tested via simulation and real datasets. A mathematical model has also been drawn for this problem. It models the cloud computing infrastructure with the ability to inter-operate among the clouds. It assumes the virtual machine as the smallest computational unit in cloud computing. The results have shown better job mapping to resources from the perspective of throughput that is improved by 2 8 % ,while the execution time is improved by 2 9 % and the amount of data transfer by 9 9 % . The second part concerns the bandwidth allocation in a virtualized environment for impatient jobs. In this part, we address the problem of immediate jobs that have huge amount of data and the ability to improve the resource allocation to meet the job dead-lines. W e modeled the problem as a set of events and proposed an algorithm that finds a proper virtual machine that can donate its bandwidth amount with full compliance to virtual machines deadlines and Q uality of Service ( Q oS) constraints. The proposed algorithm was transplanted in an adopted scheduling algorithm and tested using simu-lation with a synthetic dataset. The simulation results showed better throughput with dynamic BW allocation by 2 1 .1 % than static allocation due to better resource allo-cation. Furthermore, the algorithm showed 1 0 .0 7 % better bandwidth utilization in a virtualized environment. The third part looks at the negotiation process of the Service L evel Agreement (SL A) under cloud computing. Previously proposed models in literature have many steps for conformation which consume precious time for impatient jobs. We proposed a model for SL A negotiation which has the ability of offer-bid counter and rapid assigning in an immediate mode. System finite automata and control flow have been drawn. The proposed system is evaluated via simulation using synthetic data. F rom the results, the proposed algorithm improved the jobs throughput by reducing jobs waiting time by 8 1 .5 % , allowing more jobs to meet their deadlines