AI Enabled Resources Scheduling in Cloud Paradigm

Cloud Computing was evolved as one of the paradigm, which gives services to users in a utility-based manner. Services of cloud computing were extended to various fields and applications. Due to the enormous number of users, flexibility and easy to use nature of cloud paradigm, many of companies are trying to migrate towards cloud paradigm but from a cloud provider perspective it is a difficult to job to handle or schedule these heterogeneous workloads, which are coming onto cloud console. Therefore, it is important for a cloud provider to employ a task scheduling mechanism, which should be more proactive based on the nature of workloads coming onto cloud interface and how effectively they are scheduled onto suitable virtual resources. Many of existing scheduling algorithms used nature or bio inspired techniques to model schedulers as scheduling problem in cloud paradigm is a classical NP-Hard problem but still to make a schedule for a task onto a suitable VM based on its processing capacity while minimizing its makespan, energy consumption and other operational costs is still a tedious job as incoming user requests are highly dynamic in nature. In this paper, we have used a deep reinforcement learning technique i.e. DDQN model to make decisions of scheduling in cloud paradigm while checking incoming requests and underlying resources for every task. Task priorities are evaluated for all incoming tasks and prioritized tasks are fed to our scheduler and based on imposed conditions our scheduler will make decisions effectively. This entire research implemented on cloudsim. Extensive simulations are conducted by generating workload randomly and from realtime workload traces. Finally, our proposed scheduler is evaluated against existing baseline approaches i.e. Round Robin, FCFS, and Earliest Deadline first. From Simulation results, our proposed approach shown a huge impact over existing baseline approaches in terms of makespan, Energy consumption.