CS 4740 @ UVA - Cloud Computing Final Project

AI-Driven Multi-Cloud Serverless Scheduler

This repository explores reinforcement-learning–based scheduling of serverless functions across AWS Lambda, Google Cloud Functions, and Azure Functions. The scheduler learns which cloud to invoke (and when) to minimise end-to-end latency, energy consumption, and cost.

---

Features

• Q-learning agent that adapts to workload patterns.
• Baseline Round-Robin and Least-Connection schedulers for comparison.
• Unified wrapper for invoking each provider’s serverless API.
• CSV + console metrics for latency, energy-use (mocked), and dollars spent.

---

Quick Start

1 — Install

git clone https://github.com/jpeetla/cloud_computing.git
cd cloud_computing
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

2 — Export Cloud Credentials

Set the following environment variables before running the training script:

Provider	Required Variables
AWS	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION, AWS_LAMBDA_NAME
GCP	GCP_SA_JSON_PATH, GCP_PROJECT, GCP_FUNC_NAME
Azure	AZURE_SUB_ID, AZURE_RG, AZURE_FUNC_APP

💡 Tip: create a .env file with all variables and run source .env to quickly load them.

---

3 — Train & Evaluate

To train the reinforcement learning agent and evaluate its performance:

python train.py --episodes 200 --max_steps 100 --decay 0.99

This script performs three actions:

• Trains the agent using Q-learning and saves the learned Q-table (qtable.pkl).
• Benchmarks the Q-learning scheduler against:
  • Round-Robin scheduler
  • Least-Connection scheduler
• Outputs performance metrics (latency, cost, energy) to evaluation_summary.csv.

---

Configuration

Modify hyperparameters in config.py as needed:

Parameter	Value	Description
ALPHA	0.05	Learning rate
GAMMA	0.95	Discount factor
EPSILON	0.20	Exploration rate

---

Final cloud computing project by Jayanth Peetla and Varun Pavuloori