Task Orchestration Simulator

This package is a discrete-event simulator designed to compare task scheduling latency across three different architectures.

Modeled Strategies

1. Centralized (Kubernetes-like)

• A central scheduler directly assigns tasks to nodes.
• The central scheduler monitors node liveness directly.
• If a node failure is detected during a direct probe immediately before scheduling, a reschedule occurs, increasing latency.

2. Blockchain

• Assumes a scheduler is elected via a blockchain network.
• Node liveness is reflected in the ledger, incurring overhead for election, consensus, and verification.
• If the ledger is stale, tasks may be incorrectly assigned to failed nodes, leading to subsequent reschedules.

3. Mutual Cloud (Our Proposed Architecture)

• A control server receives task submissions and manages multi-queues.
• Execution agents pull tasks from the control server.
• Management agents on each node monitor execution agents and reflect liveness quickly via a Distributed Hash Table (DHT).
• Liveness check costs are low due to DHT lookups and multi-queue dispatching, resulting in relatively lower modeled reschedule overhead.

Usage Examples

usage: main.py [-h] [--num-tasks NUM_TASKS] [--num-nodes NUM_NODES] [--failure-rate FAILURE_RATE] [--arrival-rate ARRIVAL_RATE]
               [--mean-task-duration MEAN_TASK_DURATION] [--node-repair-time NODE_REPAIR_TIME] [--max-retries MAX_RETRIES] [--seed SEED]
               [--mode {centralized,blockchain,mutual_cloud,mutual,all}] [--output-csv OUTPUT_CSV]

Scheduling latency simulator for centralized, blockchain, and mutual cloud modes

options:
  -h, --help            show this help message and exit
  --num-tasks NUM_TASKS
  --num-nodes NUM_NODES
  --failure-rate FAILURE_RATE
  --arrival-rate ARRIVAL_RATE
  --mean-task-duration MEAN_TASK_DURATION
  --node-repair-time NODE_REPAIR_TIME
  --max-retries MAX_RETRIES
  --seed SEED
  --mode {centralized,blockchain,mutual_cloud,mutual,all}
  --output-csv OUTPUT_CSV

Run simulation:

python -m main --num-tasks 2000 --num-nodes 64 --failure-rate 0.05 --mode all