Fault-Tolerant Parallel Raytracer

Abstract

We present a hierarchical three-tier master-slave architecture for distributed raytracing designed to support a large, volatile population of participating compute clients while maintaining application-level fault tolerance and minimal scene-data replication.

Our system partitions task scheduling, scene geometry distribution, and tile aggregation across three tiers so that:

• Arbitrary Worker Failures: The cluster survives any number or pattern of client node disconnects without corrupting or stalling the rendered frame.
• Mid-Tier Server Failover: Non-concurrent intermediate server failures are automatically detected and re-routed.
• Near-Linear Scaling: Live cluster benchmarks across 50+ lab machines—including a live demonstration where machines were randomly unplugged from the network mid-render to simulate cascading hardware faults—demonstrated resilient recovery and near-linear rendering throughput scaling.