Energy- and Thermal-Aware Resource Management for Heterogeneous Edge–Cloud Systems with Workload-Dependent Capacity Degradation
Description
This repository contains the fully specified reference configuration, retained numerical outputs, corrected statistical summaries, verification code, and figure-source files supporting the study “Energy- and Thermal-Aware Resource Management for Heterogeneous Edge–Cloud Systems with Workload-Dependent Capacity Degradation.” The study examines heterogeneous edge–cloud systems in which current workload affects subsequent service capability through thermal degradation and cooling. The RCV-1 reference configuration combines finite-batch BMAP arrivals, Erlang-2 phase-type service, heterogeneous high-performance and energy-efficient edge servers, first-order RC thermal dynamics, hysteretic cooling, thermal-aware admission, and elastic cloud fallback. The package includes: (1) dual-scale 1H+1E fine-grid CTMC versus continuous-RC Hybrid-DES validation; (2) the 2H+2E workload–thermal campaign; (3) M0/M1/M2 thermal-model simplification and policy screening; (4) equal-mean Poisson, Moderate, and Strong traffic families; (5) a complete 63-policy enumeration with marginal and policy-family screens; (6) heterogeneous-server composition comparisons and architecture-specific searches; (7) cloud-energy crossover bootstrap intervals; and (8) high-resolution publication figures and their source data. Version 3 is the controlled corrected release aligned with the audited manuscript. It replaces the previous archive with corrected asymptotic customer-count dispersion indices for the Moderate and Strong BMAPs, exact continuous-time integration of fractional nominal-capacity erosion and the 75°C risk-reference fraction, threshold-split thermal-regime accounting, a source-derived six-candidate M0 selection procedure, and a joint familywise screen covering all 315 architecture–policy combinations. Under the declared familywise rule, policies 53 and 54 form the empirical sample-mean Pareto set, and only the 2H2E architecture has policies passing the joint architecture–policy screen among the five tested compositions. Independent verification code reconstructs the reported summaries, validation comparisons, traffic descriptors, erosion identities, policy selections, feasibility decisions, cloud-energy crossover values, Pareto membership, architecture diagnostics, figure-source values, and SHA-256 manifest. All 750 corrected archived-result verification checks pass with zero failures. No external empirical dataset is used. The experiments are based on analytically specified stochastic models and synthetic, mechanism-oriented numerical configurations. The archive supports verification from retained outputs and provides executable code for the workload–thermal campaign. It does not claim platform-specific thermal calibration or provide a single end-to-end simulator for regenerating every experiment.
Files
Steps to reproduce
Download and extract SUSCOM_Reproducibility_Data.zip. Install Python 3.10 or later and the required packages using: python -m pip install -r 09_VERIFICATION_CODE/requirements.txt. From the extracted package root directory, run: python 09_VERIFICATION_CODE/verify_archived_results.py. Confirm that the final output reports: PASS checks: 750; FAIL checks: 0; ALL CORRECTED ARCHIVED-RESULTS VERIFICATION CHECKS PASSED. To regenerate the corrected derived files and Figure 5 before verification, run the following commands in order: python 09_VERIFICATION_CODE/recompute_corrected_analysis.py; python 09_VERIFICATION_CODE/regenerate_figure_5.py; python 09_VERIFICATION_CODE/verify_archived_results.py. To rerun the complete nine-cell workload–thermal campaign with the archived seed map, run: python 03_WORKLOAD_THERMAL/Workload_Thermal_Run_Script.py --output-dir regenerated_workload_thermal --workers 4. The archive supports full regeneration of the workload–thermal campaign and verification of the other reported numerical blocks from retained outputs. It does not provide a single end-to-end raw simulator for regenerating every experiment.
Institutions
- Sangji UniversityGangwon-do, Wŏnju