.png){kind=icon}
.png){kind=icon}
Return
A Package-Aware Approach for Function Scheduling in Serverless Computing Environments
.png){kind=icon}
PRE
AI
DOI：10.1007/s10723-023-09657-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Cloud computing is a computing model, that offers scalable, cost-efficient computing resources based on a pay-as-you-go model to its users. The serverless computing model also known as Function as a Service enables users to run their code as cloud functions without worrying about managing the underlying infrastructure and it would cost less as well. These advantages encouraged developers to deploy and execute their applications in the cloud. One of the most important challenges in cloud environments that this research aims to tackle is scheduling. Scheduling algorithms are used to map incoming requests to computational resources in a way to fulfill one or more goals. In this research, we have leveraged the SARSA reinforcement learning algorithm and proposed SFSchlr. SFSchlr is a function scheduling algorithm that can be used in a Function as a Service platform. SFSchlr performs the learning operation online and is aware of the data dependency of the functions on each other as well as their required packages and libraries to decrease function execution turnaround time. A monitoring strategy is also introduced that runs alongside the scheduler. It manages computing resources and prevents saturation of existing workers by scaling up or out their resources. To decrease the overall cost and increase resource availability, it releases the acquired computational resources by removing inactive workers. Finally, we have implemented and evaluated the SFSchlr by comparing it with two open-source scheduling algorithms as well as a state-of-the-art dependency-aware scheduler. We have found that the proposed algorithm demonstrates up to 58% improvement in function turnaround time and up to 69.5% improvement in resource utilization.
Keywords:
Cloud computing
Scheduling
Function as a service
Reinforcement learning
Serverless computing
Journal
IF:
2.9
Papers:
759
Citations:
1.2K