Thesis research plan

Performance Comparison of Serverless Function Orchestration Tools for Small-Scale Web Application Backends

Serverless computing abstracts much of the underlying infrastructure management for developers building small-scale web application backends. Orchestrating multiple discrete serverless functions into cohesive workflows requires dedicated tools, with major cloud providers each offering native solutions. Because execution models, resource allocation, and provider-specific behavior differ, a comparison must carefully define equivalent test conditions and interpret results within each service's documented constraints. This work proposes a controlled comparison of three provider-native tools under standardized conditions to clarify performance tradeoffs for small-scale deployments.

Computer ScienceBachelor levelReviewed by Topic Production Automation Updated 2026-08-20 01:45:09
01

Research questions

Use these questions to keep the study focused and to connect each chapter to evidence you can realistically collect.

  1. How do end-to-end orchestration latencies differ between AWS Step Functions, Azure Durable Functions, and Google Cloud Workflows for small workflow patterns?
  2. What differences in throughput under bounded concurrency exist between the three tools when running equivalent small-scale workflows?
  3. How do warm vs cold start behavior and failure recovery characteristics compare across the three orchestration tools?
  4. To what extent do provider-specific quotas, execution semantics, and billing models limit direct performance comparability for this use case?
02

Methodology and data plan

Suggested approach

This study will implement two or three identical small orchestration patterns on each platform, matching workflow logic, payload sizes, retry policies, regional placement, and observation windows across all tools. Before testing, the protocol will fix the trial count, concurrency levels, runtime versions, regions, execution order, timing boundaries, and operational definitions for warm and cold runs. Repeated trials will capture natural variation in performance. Controlled testing will measure bounded-concurrency performance and recovery after a single injected activity failure. A prespecified test budget and stop rule will limit financial exposure, and the researcher must verify account billing controls before execution.

Data and sample

Raw timestamps for end-to-end workflow execution, per-step execution durations, concurrency status, and failure recovery times will be collected for all trials. Configuration snapshots for each deployment will be retained to document equivalence of test conditions. Summary statistics will be generated for full latency and throughput distributions, and uncertainty estimates will be reported alongside central tendency measures. All raw data and configuration will be stored in a version-controlled repository to support reproducibility.

03

Suggested thesis outline

Adapt the sequence to your institution’s format and the evidence available to you.

  1. 01

    Background and Related Work

    Introduce serverless function orchestration, explain core concepts, and contextualize the need for controlled performance comparisons for small-scale use cases.

  2. 02

    Platform and Test Pattern Design

    Document the three selected orchestration tools, the small workflow patterns chosen, and the steps taken to create equivalent test configurations across platforms.

  3. 03

    Experimental Methodology

    Describe the test environment, measurement approach, controlled variables, and repeated trial design used for performance data collection.

  4. 04

    Data Analysis Plan

    Outline the approach to calculating and reporting latency, throughput, warm/cold start behavior, and failure recovery metrics, including distribution and uncertainty reporting.

  5. 05

    Planned Analysis and Discussion

    Frame how performance tradeoffs will be discussed in context of provider constraints, with no universal ranking of tools based on limited testing.

  6. 06

    Conclusion and Future Work

    Summarize the planned contribution, highlight key limitations, and propose directions for future broader comparisons of serverless orchestration tools.

04

Proposed abstract

Planning draft — not completed findings

Directly comparing serverless function orchestration tools requires a controlled design because provider execution semantics and service limits differ. This work proposes a side-by-side performance comparison of AWS Step Functions, Azure Durable Functions, and Google Cloud Workflows, tested on identical small orchestration patterns with matched configuration for workflow logic, payload size, retry policies, and regional placement. The study will measure end-to-end orchestration latency, throughput under bounded concurrency, observable warm-versus-cold behavior, and recovery time after a controlled activity failure across repeated trials. Analysis will report full performance distributions and uncertainty rather than single point estimates, and will interpret differences in light of provider quotas, execution semantics, telemetry, and billing constraints. The intended contribution is a reproducible framework for discussing tool tradeoffs within the defined small-backend test conditions, not a universal ranking.

05

Evidence and search directions

  • Consult the AWS Step Functions documentation on workflow types to correctly categorize and deploy the test workflow per service specifications
  • Use Azure Durable Functions performance and scaling guidance to align test deployment with expected service behavior
  • Refer to Google Cloud Workflows documentation to confirm core capabilities and service execution model for test design
  • Reference Google Cloud Workflows quota and limit documentation to plan test trial counts and concurrency bounds that stay within service limits
  • Use NIST SP 800-145 as a foundational definition for the cloud-service terminology used in this study

Starting sources and databases