Optimizing serverless computing: A comparative analysis of multi-output regression models for predictive function invocations

Mustafa Daraghmeh

doi:10.1016/j.simpat.2024.102925

2024-07-01

Mustafa Daraghmeh *

Anjali Agarwal

Yaser Jararweh

DOI：10.1016/j.simpat.2024.102925

原文链接

查看原文

摘要

En 中文

In the rapidly evolving domain of serverless computing, the need for efficient and accurate predictive methods of function invocation becomes paramount. This study introduces a comprehensive suite of innovations to improve the predictability and efficiency of function invocation within serverless architectures. By employing multi -output regression models, we perform a multi-level analysis of function invocation patterns across user, application, and function levels, revealing insights into granular workload behaviors. We rigorously investigate the impact of windowing techniques and dimensionality reduction on model performance via Principal Component Analysis (PCA), offering a nuanced understanding of data complexities and computational implications. Our novel comparative analysis framework meticulously evaluates the performance of these methods against various windowing configurations, utilizing the Azure Functions dataset for real -world applicability. In addition, we assess the temporal stability of the models and the variation of day-to-day performance, providing a holistic view of their operational viability. Our contributions address critical gaps in the predictive modeling of serverless computing and set a new benchmark for operational efficiency and data -driven decision-making in cloud environments. This study is poised to guide future advancements in serverless computing, driving theoretically sound and practically viable innovations.

Keyword:

Serverless computing

Predictive function invocation

Multi-output regression models

Windowing techniques

Comparative analysis framework