TriSchedRL

A Guarded Deep Reinforcement Learning Framework for SLA Latency and Energy-Aware IoT Task Scheduling in Edge–Cloud Environments

---

Overview

TriSchedRL is a research-driven implementation of a Guarded Deep Reinforcement Learning (DRL) framework designed to optimize SLA latency, energy consumption, and resource utilization for IoT task scheduling across heterogeneous edge–cloud infrastructures.

This repository contains the complete modular implementation, experimental pipeline, evaluation scripts, and reproducible configurations used in the associated SCI research paper.

---

Key Features

• Guarded Deep Reinforcement Learning Scheduler
• SLA-Aware Policy Learning
• Energy and Latency Joint Optimization
• Edge–Cloud Continuum Simulation
• Modular and Reproducible Research Pipeline
• Ready for Academic Benchmarking and Extensions

---

Research Motivation

Modern IoT environments generate heterogeneous workloads requiring real-time processing. Traditional schedulers struggle to:

• Guarantee SLA constraints
• Adapt to dynamic network states
• Balance energy vs. latency trade-offs

TriSchedRL introduces:

Innovation	Description
Guard Module	Prevents unsafe scheduling decisions
Tri-Objective Optimization	Latency + Energy + Reliability
Adaptive RL Policy	Learns optimal offloading and scheduling behavior

---

System Architecture

Module	Description
src/env	Edge–cloud environment simulation
src/agents	Actor-critic policy network and replay buffer
src/guard	SLA guard with feasibility checking and repair
src/meta	Meta-controller for adaptive policy signals
src/predictors	Latency, energy, and SLA risk predictors
src/baselines	Baseline scheduling algorithms for comparison
src/evaluation	Metrics logging and performance evaluation
src/workloads	IoT task generation and workload modeling

---

Repository Structure

TriSchedRL/
│
├── README.md
├── requirements.txt
│
├── configs/
│   ├── agent.yaml
│   ├── env.yaml
│   ├── meta.yaml
│   └── workload.yaml
│
├── scripts/
│   ├── train_trischedrl.py
│   ├── evaluate_and_plot.py
│   ├── run_baselines.py
│   ├── run_all.sh
│   └── smoke_test_guard.py
│
├── src/
│   ├── agents/
│   │   ├── actor_critic.py
│   │   ├── networks.py
│   │   ├── replay_buffer.py
│   │   ├── state_vectorizer.py
│   │   └── __init__.py
│   │
│   ├── baselines/
│   │   ├── eft.py
│   │   ├── fixed_weight_sum.py
│   │   ├── least_energy.py
│   │   ├── minmin_maxmin.py
│   │   └── __init__.py
│   │
│   ├── env/
│   │   ├── edgecloud_env.py
│   │   ├── network.py
│   │   ├── resources.py
│   │   ├── sla.py
│   │   └── __init__.py
│   │
│   ├── evaluation/
│   │   ├── logger.py
│   │   ├── metrics.py
│   │   └── __init__.py
│   │
│   ├── guard/
│   │   ├── feasibility.py
│   │   ├── fallback.py
│   │   ├── repair.py
│   │   └── __init__.py
│   │
│   ├── meta/
│   │   ├── meta_controller.py
│   │   ├── signals.py
│   │   └── __init__.py
│   │
│   ├── predictors/
│   │   ├── aggregator.py
│   │   ├── energy_predictor.py
│   │   ├── feature_builder.py
│   │   ├── latency_predictor.py
│   │   ├── sla_risk_predictor.py
│   │   └── __init__.py
│   │
│   └── workloads/
│       ├── generators.py
│       ├── task.py
│       └── __init__.py
│
└── tests/
    ├── test_env_step.py
    └── test_guard.py

---

Installation

Step 1 — Clone Repository

git clone https://github.com/<your-username>/TriSchedRL.git
cd TriSchedRL

Step 2 — Create Environment

conda create -n trischedrl python=3.10
conda activate trischedrl

Step 3 — Install Dependencies

pip install -r requirements.txt

---

Running the Framework

Train the Guarded RL Scheduler

python scripts/train_trischedrl.py

Evaluate and Plot Results

python scripts/evaluate_and_plot.py

Run Baseline Comparisons

python scripts/run_baselines.py

Run Full Pipeline

bash scripts/run_all.sh

Smoke Test the Guard Module

python scripts/smoke_test_guard.py

---

Configuration

All hyperparameters and settings are managed via YAML files in the configs/ directory:

File	Purpose
agent.yaml	RL agent hyperparameters (learning rate, batch size, etc.)
env.yaml	Edge–cloud environment settings
meta.yaml	Meta-controller configuration
workload.yaml	IoT workload generation parameters

---

Evaluation Metrics

Metric	Description
SLA Latency Violation Rate	Percentage of tasks violating SLA deadlines
Energy Consumption	Total energy used across edge–cloud nodes
Task Completion Time	End-to-end task processing latency
Throughput	Number of tasks completed per time unit
Resource Utilization	CPU/memory usage across nodes
Reward Convergence	RL training stability over episodes

---

Baselines

Baseline	File	Description
EFT	eft.py	Earliest Finish Time heuristic
Min-Min / Max-Min	minmin_maxmin.py	Classic task-to-resource mapping
Least Energy	least_energy.py	Energy-greedy assignment
Fixed Weight Sum	fixed_weight_sum.py	Static multi-objective weighted sum

---

Testing

python -m pytest tests/

Test	Coverage
test_env_step.py	Environment step logic and state transitions
test_guard.py	Guard module feasibility and repair checks

---

Requirements

Python >= 3.10
torch
numpy
pandas
gymnasium
matplotlib
pyyaml

---

Contributing

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Submit a pull request

---

License

This project is released under the MIT License. See LICENSE for details.

---