Modeling Performance of Microservices Systems with Growth Theory

Area of Science:

• Computer Science
• Software Engineering
• Performance Engineering

Background:

• Microservices architecture is increasingly popular but doesn't guarantee system performance.
• Understanding and predicting performance requirement violations is crucial for system resilience.
• Techniques like load balancing and scaling need data on performance degradation to be effective.

Purpose of the Study:

• To analyze microservice performance requirement violations using time series analysis.
• To develop methods for detecting resilient and non-resilient microservices.
• To predict the future performance behavior of microservices.

Main Methods:

• Modeling performance violation occurrences as a stochastic process using growth theory.
• Applying growth models to both in-vitro e-commerce and in-production telecommunication systems.
• Interpreting growth models to characterize microservice transient performance behavior.

Main Results:

• Non-linear S-shaped growth models accurately capture violations in resilient microservices, indicating limited performance degradation.
• Linear models describe non-resilient microservices with constant, unbounded performance violations.
• Analysis of real-world systems revealed additional resilience profiles showing varied service reactions.

Conclusions:

• Growth theory effectively models performance violations in both simulated and real-world microservices.
• Model calibration is computationally inexpensive.
• The derived models offer insights into performance violation trends, aiding engineers in identifying problematic microservice operations and characterizing resilience.