Mongrel of FA-CRA and CG-CRA: Self Healing Resource Allocation with Optimized Job-Scheduling in SDN-Accelerated HPC Infrastructure

Abstract

 The traditional Job Management System (JMS) handles only physical resources as computational resources. Computational resources are assigned to a job as computing nodes in HPC (High Performance Computing) cluster system and the job processes are executed directly on allocated computing nodes. Thus the prevailing method of HPC network does not considers the process of Self-healing. Self-healing mechanism includes features like self-detection, self-repairing and self-configuring in the infrastructure of network towards maximising the reliability, resilience, safety and availability of the HPC network. Thus in our framework by considering a self-healing Software Defined Networking (SDN)-aware SH work, which formally analyse the trade-offs between timeliness and volume of the load information being revealed which enables the efficiency and granularity of the control input necessary to achieve fast reconfiguration which in turn enabling the throughput of the HPC network to be improved. The framework comprises the SDN accelerated HPC network for resource allocation using mongrel of FA-CRA (Fairness Aware Cooperative Resource Allocation) and CG-CRA (Coalition Game Based Resource Allocation) Algorithms and in order to make the resource allocation more effective manner an absolute Job scheduling is done by combining the priority based scheduler DLS (Dynamic Level Scheduling) and time based scheduler ETF (Earlier Time First). Thus the throughput of the system is improved. This framework can be implemented in NS₂ Simulation.