Research On Algorithms Of Minimizing Financial Cost Of Clouds Workflows Under Deadline Constraints

As a service computing model,clouds have been widely deployed and used for computing and storage in many applications.With an explosive growth of data,a single cloud-based data center is oftentimes insufficient to meet the enormous computing demands.As a result,an increasing number of applications are exploring the use of multiple clouds to perform data-and compute-intensive tasks to improve application performance and resource utilization.In particular,as cloud platforms continue to mature and proliferate,many large-scale workflows have migrated to multi-cloud environments for cost-effective data analysis.In such cloud-based workflow applications,financial cost is a major concern in addition to traditional performance requirements such as execution time.This thesis is focused on minimizing the financial cost of workflows under deadline constraints.We propose two mapping schemes for financial cost optimization,including two core algorithms,i.e.,Allocate VM Types(AVMT)and Choose PM and Bandwidth Types(CPMBT).This thesis consists of the following technical components:(1)We design an algorithm to properly allocate VM instances to execute workflow modules,which uses the shortest path and critical path to classify workflow modules and then strategically chooses suitable VM types.(2)We design an algorithm to choose physical machines for provisioning VM instances and allocate network bandwidths over links connecting different clouds,which decides the upgrade/downgrade of VM types and bandwidths by comparing the makespan with the deadline.(3)We prove that the proposed deadline-constrained workflow scheduling problem for financial cost minimization is NP-complete,and design two mapping schemes to schedule a single workflow and multiple workflows,respectively,by integrating the above two algorithms.These two mapping schemes can effectively reduce financial cost of the workflows.(4)We conduct simulation experiments and evaluate the performance of the proposed algorithms.The results show that the mapping schemes are superior to other scheduling schemes.In particular,WMFCO yields up to 49.66%,9.3% and 13.95% improvement,respectively,compared to Max-min,MCWM and DPSOGA in terms of financial costs,and MWMFCO yields up to 47.92% and 13.83% improvement compared to Max-min and DPSOGA.