On Skipping Redundant Computation via Smart Task Deployment for Faster Serverless
Mohammad Vatandoost Silab, Seyedeh Bahereh Hassanpour, Ahmad Khonsari, Aresh Dadlani · 2022
In serverless architectures, the scheduler component in the controller acts as a load-balancing entity and distributes the arriving tasks based on the availability of free resources on the machines. We argue that offering a multi-step decision-making process makes it feasible to allocate tasks to machines prudently, thus reducing the average response time to users. Aiming to achieve this, we propose to equip the controller with a limited cache for diverse purposes, including preserving the hash of task outcomes and storing the history of task deployment, in this paper. The response time is then further optimized by employing a batch-based technique and grouping the tasks. Evaluation results reveal that the proposed cache-aided serverless architecture can increase the average response time speed by nearly 21 %, on average, in the decision-making process.