An empirical study of open-loop load generators for microservices and web workloads

Mittal, Yug

Permalink

https://hdl.handle.net/2142/129222 Copy

Description

Title

An empirical study of open-loop load generators for microservices and web workloads

Author(s)

Mittal, Yug

Issue Date

2025-04-21

Director of Research (if dissertation) or Advisor (if thesis)

Gupta, Indranil

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Distributed Systems
• Workload Generators
• Load Testing
• Stress Testing
• Performance Testing
• Microservices
• Web Workloads

Abstract

Performance testing and load testing are pivotal for ensuring web application reliability and user satisfaction, so under optimal conditions, workload generators have to generate the workload desired by testers and report the latencies that are representative of the system under test. In the past, researchers and engineers have reported questionable results and behaviors with their workload generators. These questionable results raise a question: Are the system evaluations (including research papers) done using LGs accurate? This study examines common practices in generating synthetic traffic for benchmarking application latency, revealing that prevalent workload generators are vulnerable to obstacles that prevent the generation of the proper workload and skew latency measurements from those of the actual application. We present a comprehensive evaluation across multiple commonly used workload generators under varied configurations to identify pitfalls that each cannot counter. We contribute cautionary insights into latent flaws within widely used open-loop workload generators and actionable strategies for improving the accuracy of server-side performance assessments. We evaluate these widely used open-loop workload generators, their accuracy in generating the workload, how the application receives that workload without limitations on machine resources, and the reflection of the recorded latency measurements from the workload generator towards the application. Upon this analysis, we find various limitations and drawbacks with open-loop load testing that limit how the workload generator generates and how the application receives the workload. We also see how these limitations and disadvantages impact the latency measurements recorded by the LG.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129222

Copyright and License Information

Copyright 2025 Yug Mittal

Owning Collections