Synchronous vs Asynchronous I/O: Blocking, Non‑Blocking & Event‑Driven Models

I/O Types

Synchronous vs Asynchronous (synchronous, asynchronous): focusing on notification mechanism

Synchronous: After a process issues a system call, it does not receive an immediate response; when a response arrives, it is the final result.

Asynchronous: After a process issues a system call, it receives an immediate return that is not the final result; once the kernel finishes processing, it notifies the process that the call has completed.

Blocking vs Non‑Blocking (blocking, non‑blocking): focusing on the caller's state when the system call completes

Blocking: The caller remains suspended until the result is returned, then it can continue working.

Non‑Blocking: The caller is not suspended while waiting for the result; the system call itself does not block the caller.

I/O Model Classification

Blocking I/O

Non‑Blocking I/O

Multiplexed I/O

Event‑Driven I/O

Asynchronous I/O

Blocking I/O Model

When a user process initiates a system call, the process stays suspended until the kernel finishes the task (e.g., reading data from disk into kernel memory, processing it, and then transferring it to the process's memory). Only after this transfer can the process resume work.

Non‑Blocking I/O Model

After the system call, the process is not suspended; it repeatedly polls the kernel to check if the operation is complete. When the kernel finishes processing (data moved from disk to kernel memory), the process is briefly suspended while the kernel transfers the data to the process memory, after which the process can continue.

Multiplexed I/O Model

The process can issue multiple requests to the kernel; mechanisms such as select, poll, and epoll enable I/O multiplexing. Multiplexing allows a single process to monitor many descriptors and be notified when one becomes ready for reading or writing. Although select/poll/epoll are fundamentally synchronous (the process must perform the actual read/write after notification), asynchronous I/O offloads the data transfer from kernel to user space without the process’s involvement.

Event‑Driven I/O Model

When the process issues a system call, the kernel immediately returns a signal indicating receipt. The process can perform other work while the kernel prepares the data. Once ready, the kernel notifies the process, which is then suspended briefly to receive the data and can resume processing.

Asynchronous I/O Model

After the system call, the kernel returns a signal that the request has been received, allowing the process to continue other tasks. When the kernel finishes processing and transfers the data to the process memory, it notifies the process, which can then simply read the data and proceed.