Timmer, Rick (2024) Cost-Effective Machine Learning Inference with AWS Lambda: Evaluating Serverless Resource Configurations. Master's Thesis / Essay, Computing Science.
|
Text
mCS2024TimmerR.pdf Download (2MB) | Preview |
|
![]() |
Text
toestemming.pdf Restricted to Registered users only Download (133kB) |
Abstract
In cloud computing, serverless offerings like AWS Lambda offer notable benefits in scalability and resource management. In theory, the flexibility and auto-scaling features of serverless platforms align well with machine learning (ML) demands, allowing computational resources to be dynamically allocated based on the real-time requirements of ML models. In order to optimize the efficiency and affordability of machine learning inference tasks, it is crucial to have an understanding of the various resource configurations and their implications. This thesis explores the performance and cost of different AWS Lambda resource setups for running ML inference workloads. The study examines how memory allocation and concurrency settings affect computational efficiency and costs by conducting systematic experiments with various ML algorithms—unsupervised, supervised, and large language models. The results reveal significant differences in cost and performance across various configurations. For instance, allocating 1024 MB of memory often provides a good balance between cost and performance for unsupervised and supervised algorithms. In contrast, large language models are not able to run efficiently on AWS Lambda due to significant latency and high costs, making them unsuitable for real-time applications. In terms of implications, this research provides insights into the trade-offs between computational resources and execution costs, helping stakeholders make informed decisions that ba...
Item Type: | Thesis (Master's Thesis / Essay) |
---|---|
Supervisor name: | Andrikopoulos, V. |
Degree programme: | Computing Science |
Thesis type: | Master's Thesis / Essay |
Language: | English |
Date Deposited: | 01 Aug 2024 10:19 |
Last Modified: | 01 Aug 2024 10:19 |
URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/33792 |
Actions (login required)
![]() |
View Item |