Tools
Tools
Sathiyamoorthy, Hariharan (2024) Mitigating Serverless Cold Start Latency with Predictive Function Invocation and Adaptive Caching. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
The serverless industry is estimated to be worth over $21.9 billion in 2024 due to its separation of infrastructure management from application, and at the same time, it provides scalability and immense cost reduction. There are few problems when it comes to serverless that is cold start latency, which hinders the performance. The purpose of the study is to create a hybrid approach by combining predictive modeling to predict the function invocation along with adaptive caching to reduce the cold start frequency and latency. The study involved setting up the OpenWhisk serverless platform on a cloud machine, simulating real-world serverless behavior, using machine learning techniques to predict the function invocation time, creating a robust cache algorithm, and combining these two by creating a framework called SmartFaaS, followed by comparing it with the vanilla OpenWhisk solution for benchmarking. The key finding from the study was that SmartFaaS consistently used 30% of resource throughout all the experiments and the overall execution time were maintained under 200 ms compared to OpenWhisk 1200 ms showing an approximate 83% reduction in latency. By Using the predictive model in SmartFaaS only 37% of cold starts were recorded showing efficient reduction in cold start frequency. From these results, SmartFaaS offers a more effective and cost-efficient solution for mitigating cold start issues in serverless computing. The research contributes to the current state of the art by demonstrating the effectiveness of combining predictive modeling with adaptive caching. In practice, this means developers and cloud service providers can achieve better performance and resource utilization. However, the limitations of this study were that a single node application was tested, and as of now, only Node.js applications can utilize SmartFaaS. Future work could involve testing SmartFaaS with diverse workloads and exploring other machine learning models.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Makki, Ahmed UNSPECIFIED |
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Cloud Computing |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 16 Jul 2025 11:54 |
| Last Modified: | 16 Jul 2025 11:54 |
| URI: | https://norma.ncirl.ie/id/eprint/8151 |
Actions (login required)
 |
View Item |