Performance Antipatterns: Angel or Devil for Power Consumption?

TL;DR

This study probes whether established performance antipatterns in microservices also drive higher energy consumption. It implements ten Smith & Williams antipatterns as isolated containerized services and conducts extensive load tests with synchronized performance and power measurements, plus resource utilization data. The findings reveal that degraded response times do not universally translate to increased energy use; some antipatterns show clear energy-performance coupling while others mainly incur energy costs through sustained saturation rather than latency. The work provides a systematic foundation for identifying energy-relevant antipatterns and offers replication data to support reproducibility and further exploration of energy-aware microservice design.

Abstract

Performance antipatterns are known to degrade the responsiveness of microservice-based systems, but their impact on energy consumption remains largely unexplored. This paper empirically investigates whether widely studied performance antipatterns defined by Smith and Williams also negatively influence power usage. We implement ten antipatterns as isolated microservices and evaluate them under controlled load conditions, collecting synchronized measurements of performance, CPU and DRAM power consumption, and resource utilization across 30 repeated runs per antipattern. The results show that while all antipatterns degrade performance as expected, only a subset exhibit a statistically significant relationship between response time and increased power consumption. Specifically, several antipatterns reach CPU saturation, capping power draw regardless of rising response time, whereas others (\eg Unnecessary Processing, The Ramp) demonstrate energy-performance coupling indicative of inefficiency. Our results show that, while all injected performance antipatterns increase response time as expected, only a subset also behaves as clear energy antipatterns, with several cases reaching a nearly constant CPU power level where additional slowdowns mainly translate into longer execution time rather than higher instantaneous power consumption. The study provides a systematic foundation for identifying performance antipatterns that also behave as energy antipatterns and offers actionable insights for designing more energy-efficient microservices architectures.

Figures (4)

• Figure 1: Research Process (in BPMN bpmn notation).
• Figure 2: Container diagram of the experimental infrastructure (C4 notation brown2026). Blue containers indicate data collection and experiment coordination containers; green containers indicate data collection tools; orange containers indicate the software under test (SUT); and violet containers indicate external systems.
• Figure 3: Trace Visualization of Unnecessary Processing, Circuitous Treasure Hunt, and The Ramp.
• Figure 4: Response time, CPU and DRAM power box plots across all 30 runs for every studied antipattern; boxes show median/IQR, whiskers show central spread, mean marked (outliers hidden for clarity).