A Heuristic Meta Scheduler for Optimal Resource Utilization and Improved QoS in Cloud Computing Environment

2. Problem Definition

In cloud IaaS model, service consumer specify hardware and software configuration of Virtual Machines(VM) to be created and maps a set of jobs to each VM. On a broader classification, the nature of the submitted jobs are parallelizable, non-pre-emptive, independent and time-critical. It is expected that the VMs complete the processing of jobs within deadline, once VMs are launched on the appropriate host. Each VM processes an array of jobs associated with an Expected Completion Time (ECT). The speed with which the VM processes the jobs depends on the speed (MIPS) of the Processing Elements (PE) on which VM is launched and also allocation policy assigned with the host. The objective of the Meta scheduler is to identify the suitable hosts for launching VMs as per the customer requirements.

In the first few cycles of resource allocation as the cloud is initiated, the resources of the host(servers) are utilized properly. In due course, few of the jobs fired by the customer are completed and the corresponding VMs are destroyed. This leads to creation of resource holes in the servers. A VM created with n PEs from the same host outperforms a VM that is created by n PEs across several hosts in the cloud, because the latter suffer from increased internal latency. Hence such a VM creation will be a poor strategy of the scheduler. Sometimes the scheduler is not able to create a VM even when there are enough resources but scattered within the cloud, resulting in failure of VM creation. A new request from the customer in this situation can be put on hold or rejected. Hence, a Cloud system highly demands for a scheduler which can effectively eliminate resource holes, and minimize failure in VM creation.