Comparing BIO, NIO, and Asynchronous Models Using a Bank Process Analogy

Assume a bank with ten employees and a four‑step service process: customer fills a form (5 min), employee reviews (1 min), employee calls security to fetch cash (3 min), and employee prints a receipt and returns cash (1 min). The total time per customer is 10 minutes.

1. BIO (Blocking I/O) model : Each arriving customer is handled by a single employee who performs all four steps. With ten employees, the bank can serve 6 customers per hour per employee, i.e., 60 customers per hour total. Many employees are idle during the waiting phases, indicating low resource utilization.

2. NIO (Non‑blocking I/O) model : One employee (A) only collects forms and distributes completed forms to the remaining nine employees for the subsequent steps. The nine employees each process a customer in 5 minutes, yielding 108 customers per hour. This model introduces a thread‑pool‑like separation of concerns, similar to a mainReactor accepting connections and subReactor threads handling I/O, while worker threads perform business logic.

3. Asynchronous model : A dedicated employee (B) handles the third step (security interaction). When the cash is ready, B notifies the customer, allowing the teller to immediately proceed to the final step. This reduces the teller’s wait time to 2 minutes per customer, enabling eight tellers to serve 240 customers per hour. The pattern mirrors asynchronous RPC or HTTP calls where long‑running operations are off‑loaded, e.g., Jetty Continuations.

All three approaches demonstrate the "divide‑and‑conquer" principle: breaking a request into independent tasks handled by specialized workers dramatically increases system throughput, a concept applicable both in software engineering and broader societal workflows.