Journal
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
Volume 34, Issue 1, Pages 180-194Publisher
IEEE COMPUTER SOC
DOI: 10.1109/TPDS.2022.3214783
Keywords
Costs; Optimization; Time factors; Firing; Analytical models; Computational modeling; Cloud computing; Cloud serverless computing; performance modeling; performance optimization; cost modeling; cost optimization
Ask authors/readers for more resources
This study fills the gap in predictability and performance-cost trade-off of Function-as-a-Service (FaaS) applications by proposing formal performance and cost modeling and optimization algorithms. The proposed model and algorithms enable accurate prediction and fine-grained control over the performance and cost of FaaS applications, helping developers make informed decisions.
Function-as-a-Service (FaaS) has become a mainstream cloud computing paradigm for developers to build cloud-native applications in recent years. By taking advantage of serverless architecture, FaaS applications bring many desirable benefits, including built-in scalability, high availability, and improved cost-effectiveness. However, predictability and trade-off of performance and cost are still key pitfalls for FaaS applications due to poor infrastructure transparency and lack of performance and cost models that fit the new paradigm. In this study, we therefore fill this gap by proposing formal performance and cost modeling and optimization algorithms, which enable accurate prediction and fine-grained control over the performance and cost of FaaS applications. The proposed model and algorithms provide better predictability and trade-off of performance and cost for FaaS applications, which help developers to make informed decisions on cost reduction, performance improvement, and configuration optimization. We validate the proposed model and algorithms via extensive experiments on AWS. We show that the modeling algorithms can accurately estimate critical metrics, including response time, cost, exit status, and their distributions, regardless of the complexity and scale of the application workflow. Also, the depth-first bottleneck alleviation algorithm for trade-off analysis can effectively solve two optimization problems with fine-grained constraints.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available