Performance Predictors Project

This repository contains components for managing and evaluating Round-Trip Time (RTT) and variability predictors used in edge computing experiments. The system is modular, with separate containers for managing predictors, running individual prediction services, and post-processing experimental results.

Repository Structure

prediction_manager

• Purpose:
  Hosts the predictor manager responsible for orchestrating both RTT and variability predictors.

• Container:
  Runs as a standalone Docker container.

• Execution:
  Built using the provided Dockerfile and launches src/main.py.

rtt_predictor

• Purpose:
  Contains the RTT prediction processes.

• Container:
  Runs independently in a Docker container.

• Execution:
  Built using the provided Dockerfile and launches src/main.py.

variability_predictor

• Purpose:
  Contains the variability prediction processes.

• Container:
  Runs independently in a Docker container.

• Execution:
  Built using the provided Dockerfile and launches src/main.py.

postprocessing

• Purpose:
  Provides scripts for analyzing predictor performance at the end of experiments.

• Functionality:

  • Evaluates predictor accuracy, overhead, and configuration changes over the experiment duration.

  • Uses raw data and stores results in the data/ subdirectory.

Data

• The raw experiment data and analysis results are stored in the postprocessing/data/ directory.

• This directory is structured to separate raw logs from processed results.

Relevant articles:

• RTT predictors: P. Giannakopoulos, B. van Knippenberg, C. K. Joshi, N. Calabretta, and G. Exarchakos, “Morpheus: Lightweight RTT prediction for performance-aware load balancing,” Future Generation Computer Systems, 2026, Art. no. 108452, ISSN 0167-739X, doi: 10.1016/j.future.2026.108452.
• Variability predictors: P. Giannakopoulos, B. van Knippenberg, C. K. Joshi, N. Calabretta, and G. Exarchakos, “Runtime RTT variability predictors for performance-aware scheduling in edge computing,” in Proc. 8th Conference on Cloud and Internet of Things (CIoT), London, UK, 2025, doi: 10.1109/CIoT67574.2025.11410133