Performance Tradeoffs

Parallelism is not a silver bullet.

1. Context Window Fragmentation

If you split a task into 10 sub-tasks, you now have 10 separate context windows.

• Pro: You can process more tokens in total.
• Con: Each worker lacks the "big picture" context of the others.

2. Rate Limits

Fan-out is the easiest way to hit an OpenAI 429 Too Many Requests error. If you launch 50 parallel nodes, you trigger 50 API calls instantly. You may need a "Semaphore" or "Batching" node to throttle execution.