Introduction to Operating Systems

An operating system is a collection of software that manages the computer's hardware and software resources and provides common services for programs. That definition has two halves and both matter. It manages resources: only one program can own the CPU at a given instant, and someone has to decide who. It provides services: your program does not know how to talk to a hard disk, so it asks the OS to read a file for it.

Think of a computer system as four layers stacked on top of each other. The slides call these the four components of a computer system, and the very first recap question asks you to spot the odd one out.

Before you can understand what the OS does, you need a picture of the machine it runs on. A modern computer has one or more CPUs and a set of device controllers (disk controller, USB controller, graphics adapter) all connected through a common bus that gives them access to shared memory. The CPUs and the devices execute concurrently, and because they all want memory, they compete for memory cycles.

Most systems use a single general-purpose processor, and many also carry special-purpose processors (a disk controller often has its own tiny CPU, a GPU is another). But the trend that matters is the rise of multiprocessor systems, also called parallel systems or tightly-coupled systems: two or more processors sharing memory and the bus.

Two more terms from the slides. A dual-core design puts two cores on one chip, so a system can be multi-chip and multicore at the same time. At the very top end, a chassis containing multiple separate systems (a blade server) is also counted as one multiprocessor system.

The slides split the OS's work into four functions (the big managers) and five other important activities (the housekeeping). Examiners love asking you to classify a task into one of these buckets, so learn the exact names.

A modern OS is interrupt driven. It does not sit in a loop checking things. It waits, and when something happens it gets woken by an interrupt, handles it, and goes back to sleep or hands the CPU to a user program. Interrupts come in two families.

Now the protection question. If any program could execute any instruction, one buggy program could overwrite the OS or turn off interrupts forever. So the CPU supports dual-mode operation. A mode bit in the hardware says whether the CPU is currently running user code or OS code, and certain privileged instructions simply refuse to execute unless the bit says kernel.

The slides list five types of operating system: simple batch, multiprogramming, multiprocessor, distributed, and real-time. For each one you need the idea, the advantages, the disadvantages and an example. We take the first two here.

Examples on the slides: payroll systems and bank statements. Both are still batch today: your bank computes interest for every account overnight in one run, with nobody sitting at a keyboard. The trade-off is high throughput but low responsiveness: the machine is always busy, but no individual user gets an answer quickly.

Multiprogramming is the next step. Sharing the processor when two or more programs reside in memory at the same time is multiprogramming. The OS keeps several jobs in memory at once; this set is a subset of the job pool on disk. It picks one job and starts executing it. When that job has to wait (say for I/O), the OS switches to another job in memory. The goal: the CPU is never idle unless there are no jobs to process.

A multiprocessor operating system runs on a system with several processors that share a common physical memory. It provides higher computing power and speed. Two things to remember about how it is organised: all processors operate under a single operating system, and the multiplicity of processors is transparent: a program does not know or care which processor it is on.

A distributed operating system is different in kind. It is a collection of autonomous but interconnected computers that communicate over a shared communication network. Each node is independent with its own memory and CPU. There is no shared physical memory; cooperation happens by sending messages. Google's search index spread across thousands of servers is the everyday example.

A real-time operating system (RTOS) is used when there are very strict time requirements: missile systems, air traffic control, robots. The time between an input arriving and the system responding is the response time, and in an RTOS it must be very small and, more importantly, predictable. A normal OS optimises for average throughput; an RTOS optimises for never missing a deadline.