Workflow Scheduling With Spot Instances In Cloud Computing

Cloud computing provides a reliable, customized and quality of service guaranteed dynamic environments to end users. Under cloud environment it is possible for users to access the applications and their associated data from anywhere at any time. Cloud com-puting provides services in three different ways e.g., SaaS, PaaS, IaaS including software, platforms and infrastructures respectively. Companies can reduce total renting costs of their computing resources by renting them from cloud service providers. Resources are offered in long term (reserved) and short term (Spot-Block and On-Demand) manners which have different pricing models with various prices and time durations. Schedul-ing workflow tasks and renting efficient resources by using different pricing models to minimize the cost under user defined parameters is a huge challenge in cloud computing.Characteristics of provisioning manners are analyzed and a mathematical model is established. To address this problem, a heuristic named Cost Saving Algorithm (CSA) is proposed that mainly contain three phases:resource and sequence initialization (RSI), scheduling construction (SC) and finally multi-sequence generation (MSG). The resources initialization process avoids the unnecessary attempts of the amounts reserved resources during the schedule construction process which saves a lot of time. Scheduling construc-tion phase first efficiently scheduled workflow tasks by using free time slot and rent the using all three different pricing models to reduce the cost of execution whilst meeting the workflow deadline. As task allocation sequence exerts a great influence on the perfor-mance of the CSA in multi-sequence generation phase a local search algorithm is proposed to adjust the sequence of task scheduling.A two-phase algorithm which uses only reserved and on-demand renting manners is constructed to elaborate the performance of proposed approach and named as aHEFT. The multi-factor analysis of variance (ANOVA) is adopted to analyze the performance of cost saving algorithm with different parameters values. Experimental results showed that the average cost of aHEFT is 14.5% higher than proposed approach.