Analysis: 10,000 independent images, each with 4 sequential stages.

Two viable approaches:

Approach 1: Master-Worker (simpler, recommended)
- PRAM model: CREW (concurrent read of task queue, exclusive write to output)
- Each worker processes one complete image (load→resize→filter→save)

task_queue = all 10000 image paths

#pragma omp parallel
{
    while (image = get_next_task(task_queue)) {  // needs synchronization
        data = load(image)
        data = resize(data)
        data = filter(data)
        save(data, output_path)
    }
}

Approach 2: Pipeline (higher throughput)
- 4 stages, each with its own thread(s)
- Different images in different stages simultaneously

Stage 1 threads: Load images → queue1
Stage 2 threads: Resize from queue1 → queue2
Stage 3 threads: Filter from queue2 → queue3
Stage 4 threads: Save from queue3

Recommendation: Master-Worker is simpler and sufficient since each image is independent. Pipeline adds complexity with inter-stage queues but can provide better throughput if stages have different costs.

PDC Midterm Prep