Performance, Cost, and Scalability in Serverless Computing: A Cloud Provider Benchmark : A Study of FaaS Platforms with SeBS

Sammanfattning

Serverless computing, particularly Function-as-a-Service (FaaS), is widely adopted for its automatic scaling, event-driven execution, and pay-per-use pricing. However, performance, scalability, and developer experience vary between providers, making it difficult to choose the most suitable platform. This thesis compares three leading FaaS providers: AWS Lambda, Google Cloud Functions, and Azure Functions. The goal is to evaluate their performance under different workloads, analyze pricing models, and assess developer usability. The problem is relevant due to the ongoing evolution of serverless platforms, which creates a need for new practical evaluations. The study uses the open-source Serverless Benchmark Suite (SeBS) to run two real-world workloads: a dynamic HTML generator and an image thumbnailer. Each application is tested under cold and warm conditions, with varying concurrency levels. Metrics such as response time, performance variability, and invocation errors are collected. In parallel, documentation quality, console usability, and pricing models are evaluated. Results show that AWS Lambda delivers the most consistent performance with low variability, making it a strong choice for predictable workloads. Azure Functions has the lowest warm invocation latency but shows more fluctuation. Google Cloud Functions performs well on cold starts but degrades under high load. Cost comparisons reveal that while all providers offer generous free tiers, their billing models differ in ways that can affect long-term cost. These findings help teams choose a serverless platform based on workload characteristics, cost sensitivity, and tolerance for latency variation.