Efficient Hybrid DDPG Task Scheduler for HPC and HTC in Cloud Environment

Abstract

Task Scheduling is a crucial challenge in cloud computing as diversified tasks come rapidly onto cloud console dynamically from heterogeneous resources which consists of different task lengths, processing capacities. Generating schedules for these type of tasks is a challenge for Cloud Service Provider(CSP). Therefore, to generate task schedules in cloud paradigm effectively by considering type of task arising to cloud console and match it with respective Virtual Machine (VM), a task scheduler is formulated by using Deep Deterministic Policy Gradient (DDPG) algorithm which is used as methodology to design scheduler. This scheduler works in three stages. In the initial stage, tasks are classified based on length and processing capacity to identify them whether they are High Performance Computing (HPC) tasks or High Throughput Computing (HTC) tasks. After classification, in the second stage, resources are to be tracked which matches the corresponding nature of tasks. Finally, in the third stage, according to the VM priorities calculated based on electricity unit cost and tasks are mapped according to the priorities to the corresponding VMs. Simulations are conducted using Cloudsim with fabricated workload distributions and realtime worklogs. Finally, our proposed Hybrid workload Deep Deterministic Policy Gradient Task scheduler(HDDPGTS) evaluated over DQN, A2C algorithms. From results, it proved that our proposed HDDPGTS significantly improved makespan, Energy consumption, scheduling overhead, scalability over baseline approaches.

Keywords

• Task scheduling
• cloud computing
• makespan
• energy consumption
• DQN
• DDPG