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

A bank with ten employees processes a customer request in four steps: fill‑in form (5 min), review (1 min), security fetches cash (3 min), and print receipt (1 min). The author uses this scenario to compare three processing models.

1. BIO (Blocking I/O) model – each arriving customer is handled entirely by a single employee who performs all four steps. With a total of 10 minutes per customer, one employee can serve six customers per hour, so ten employees handle at most 60 customers per hour, leaving many workers idle during the first step.

2. NIO (Non‑blocking I/O) model – the work is split: employee A only collects the forms and then distributes them to the remaining nine employees for the subsequent steps. Assuming employee A is saturated, the nine workers each process a customer in 5 minutes, yielding 9 × (60/5) = 108 customers per hour. This mirrors the classic NIO architecture with a main reactor, sub‑reactors, and worker threads, where each thread specializes in a specific task, eliminating idle time.

3. Asynchronous model – a third employee B is dedicated to the third step (security fetching cash). While the teller finishes step 2, B retrieves the cash, allowing the teller to immediately start the next customer. With this arrangement, a teller can finish a customer in 2 minutes, so eight active tellers process 8 × (60/2) = 240 customers per hour. The author notes that this pattern corresponds to asynchronous RPC/HTTP calls in modern web services, and mentions Jetty Continuations as a concrete implementation.

Conclusion – The “divide‑and‑conquer” principle, assigning specialized staff (or threads) to specific tasks, dramatically improves throughput and applies both to computer systems and broader societal processes.