Petri net model for resource scheduling with auto scaling in elastic cloud

Abstract

Cloud service vendors offer wide range of services in a pay per use paradigm to the customers. The aim of cloud resource management is to speed up the execution of tasks and efficient usage of computing resources. The main benefit of optimal scheduling policy is reduced makespan, energy as well as cost with minimum number of service level agreement (SLA) violations. In market oriented cloud the service vendors offer vivid variety of purchasing options and also dynamic prices to the customers. In order to incorporate these purchase-promotional offers and dynamic prices, this paper proposes a Petri net model for scheduling the workload across physical servers. Here, SLA requirements considered are CPU speed, memory, makespan and bandwidth with fewer virtual machine migrations. The experimental results indicate that the proposed system efficiently performs dynamic provisioning and elasticity in multiple public clouds with auto scaling that reduces makespan, number of SLA violations, penalty cost and maximises the profit.