Performance Evaluation of Serverless Cloud-Native Applications: FaaS vs. Container-Based in AWS
Ahmad, Hamaad (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025121636936
https://urn.fi/URN:NBN:fi:amk-2025121636936
Tiivistelmä
This thesis includes a comparative performance and cost evaluation of two serverless deployment models on Amazon Web Services (AWS): AWS Lambda (Function-as-a-Service) and AWS Fargate (container-based compute). The research study analyses the behaviour of both platforms under controlled workloads by using an identical light-weight REST API, by running tests with K6 in the presence of cold, warm and mixed load patterns. The aspects that are specifically evaluated are average latency, tail per-formance (p95), scalability behaviour and cost per 1,000 requests, as these factors have a direct impact on decisions regarding deployment in cloud-native architectures.
Results indicate that AWS Lambda is highly competitive on bursty or irregular traffic in terms of low warm start latency and has a high scalability due to it is event-driven exe-cution model. However, Lambda has higher cold-start delays as well as tail latency that varies. With a dedicated container environment, AWS Fargate is able to show more predictable and consistently lower mean and p95 latency for cold, warm and mixed workloads. Cost analysis shows that Lambda is far more cost efficient for workloads that are intermittent or at low traffic volume because, the user only pays for calls and duration of the call. In contrast, where the persistent, high throughput workload will be sustained, it is preferable to use Fargate because it provides the highest computing efficiency because it is fixed. Overall, the results bring into light the performance trade-offs between Lambda and Fargate.
Results indicate that AWS Lambda is highly competitive on bursty or irregular traffic in terms of low warm start latency and has a high scalability due to it is event-driven exe-cution model. However, Lambda has higher cold-start delays as well as tail latency that varies. With a dedicated container environment, AWS Fargate is able to show more predictable and consistently lower mean and p95 latency for cold, warm and mixed workloads. Cost analysis shows that Lambda is far more cost efficient for workloads that are intermittent or at low traffic volume because, the user only pays for calls and duration of the call. In contrast, where the persistent, high throughput workload will be sustained, it is preferable to use Fargate because it provides the highest computing efficiency because it is fixed. Overall, the results bring into light the performance trade-offs between Lambda and Fargate.