Locust- inspired meta-heuristic algorithm for optimising cloud computing performance

Cloud computing offers high computational resources at a reasonable pricelevel. This has led to a great migration of users to cloud computing from other modes of computing. Cloud computing resources are offered on a pay-as-youuse basis, allowing users to be free from maintenance costs. The cloud paradigm has arisen due to a rapid growth in applications and data sizes. Even though cloud computing servers and resources may seem unlimited, this is not true, as increased server usage leads to increased energy consumption and carbon emissions. Therefore, minimising the number of active servers in a cloud-computing set-up can significantly improve energy consumption. Additionally, reducing the number of virtual machine migrations can improve the hardware reliability of the overall cloud computing system. Another aspect that can increase user satisfaction is the scheduling of users’ tasks, as many agencies, organisations, and departments are responsible for time-critical tasks that need to be completed as soon as possible at reasonable cost. This thesis presents three significant contributions to the field of knowledge. The first contribution entails a study on server consolidation, which employs the Locust Scheduling Meta-Heuristic Algorithm (LACE). This contribution is composed of three distinct parts. The first part involves a review of prior locustinspired algorithms, while the second part concerns the adaptation of the algorithm to the cloud computing paradigm. The third part addresses the limitations of LACE algorithms, leading to the proposition of a novel meta-heuristic algorithm called the Locust-Inspired Algorithm (LIA) that can effectively map virtual machines (VMs) for efficient server consolidation. This algorithm can also be used for task scheduling. The proposed algorithm efficiently maps and achieves the objective function for server consolidation, optimising energy consumption, VM migrations, and server utilisation. To validate the effectiveness of the pro posed algorithm, it was tested via simulation using real datasets. Furthermore, a mathematical model was developed, which models the cloud computing infrastructure, capable of allocating VMs to a minimum number of servers, increasing server utilisation, and triggering necessary migrations to reduce underutilised servers. The simulation results demonstrate that the proposed algorithm outperforms existing heuristic and meta-heuristic algorithms, including the benchmarking algorithm (LACE). The proposed algorithm demonstrated a 61.8% and 81.03% reduction in energy consumption and VM migrations, respectively, compared to LACE. Additionally, the proposed algorithm exhibited superior performance compared to other state-of-the-art algorithms. The second contribution of the thesis concerns the scheduling of independent tasks, called cloudlets. In this contribution, a novel analogy of the locust-inspired algorithm is presented in the field of cloudlet scheduling. The proposed algorithm has the ability to improve cloudlet allocation to meet the objective function. The problem is modelled as a set of events that locates an appropriate VM on which to allocate the cloudlet. The proposed algorithm’s efficiency is evaluated using the CloudSim toolkit and a synthetic dataset. Results reveal that it outperforms other state-of-the-art nature-inspired algorithms such as TOPSISPSO, FUGE, ACO, and MACO, with average improvements of 55.6%, 66.9%, and 31.6% in makespan, waiting time, and resource utilisation, respectively. The third contribution arises from investigating the scheduling of dependent tasks, where most of the tasks have parents and children, and the batch of tasks is called a job. These tasks are connected together based on the model structure. The scientific workflow has an immense computational requirement, which is considered data-intensive. The LIA is considered a novel algorithm that adapts the study of locust movement behaviour from biology to job scheduling in the cloud computing environment. The proposed algorithm is used with four different workflow structures (Montage, Cybershake, Inspiral, and SHIPT) and their datasets within a range of 50, 100, and 1000 tasks. The proposed algorithm is evaluated using the WorkflowSim simulation with a real dataset. From the results, the LIA improves job allocation by reducing job makespan and cutting the cost of using resources. The job scheduling of the scientific workflow can efficiently outperform state-of-the-art competitor algorithms.

Text 112717.pdf Download (2MB)

Actions (login required)

View Item