Operating systems essentials

These chapters are a quick reference for the fundamentals questions that appear in phone screens and system design follow-ups. Each section ends with questions to answer aloud; if you cannot explain an answer in two or three sentences, study that topic further. Reviewed October 11, 2026.

Processes and threads

A process has its own virtual address space, file descriptors and security context. A thread is a schedulable unit of execution inside a process; threads share the address space and descriptors but each has its own stack and registers. Creating and switching threads is cheaper than processes, but shared memory requires synchronization.

A context switch saves one thread's registers and loads another's; switching processes also changes the page tables, which can flush the TLB.

Virtual memory

Each process sees a contiguous virtual address space divided into pages, usually 4 KiB. The MMU translates virtual pages to physical frames through page tables, caching translations in the TLB. A page fault occurs when a page is not resident: the kernel loads it from disk or allocates it. Thrashing happens when the working set exceeds physical memory and the system spends its time paging.

Scheduling

Modern general-purpose kernels use preemptive, priority-aware schedulers that aim for fairness (Linux's CFS and its successor EEVDF). Know the classic trade-offs: first-come-first-served is simple but suffers convoy effects; round robin bounds waiting time; shortest-job-first minimizes average waiting time but needs job length estimates and can starve long jobs.

I/O models

• Blocking I/O: the thread waits until data is ready. One thread per connection is simple but expensive at high concurrency.
• Non-blocking I/O with readiness notification: epoll (Linux) or kqueue (BSD/macOS) report which sockets are ready; one thread can serve thousands of connections. This underlies Node.js, nginx and Netty.
• Completion-based I/O: submit operations and receive completions (io_uring, Windows IOCP).

Questions to answer aloud

• Why can a program with more threads than cores run slower?
• What happens, step by step, when a process reads a file it has never opened?
• Why does a memory leak eventually slow the whole machine before it crashes?