Jams appear with no crash. How can a computer predict one?

The input–process–output model

• Any computer system can be described with the input–process–output (IPO) model.
• Input: data going into the system, from a keyboard, a sensor or a file.
• Process: the computer carries out operations on the data, such as calculating, comparing or sorting.
• Output: the results, shown on a screen, printed, played as sound, or sent as a signal to a device.

Sequential and parallel models

• Sequential: instructions are carried out one after another, in order. Each step starts only when the one before has finished.
• Parallel: a task is split into parts that are carried out at the same time on several processors or cores.
• Parallel is faster only when the task can be split into independent parts. A part needing an earlier result must still wait, and sharing out work adds overhead.

Multi-agent models

• In a multi-agent model, many independent agents act at the same time. Each follows its own rules and reacts to the agents around it.
• An agent can be a program, a robot in a swarm, or a simulated car, bird or person.
• No single agent is in charge. The overall pattern, such as a traffic jam, a flock or a crowd, comes from the agents' interactions.

Simulation and modelling software

• A model is a simplified representation of part of the real world, built from variables and rules (an abstraction).
• A simulation runs the model over time. Change the inputs to ask "what if?".

ENIAC took days to reprogram. Your phone swaps apps instantly. Why?

Programs live in memory

• A program is a list of instructions in binary (machine code), processed by the CPU.
• A running program's instructions and data are both held in main memory (RAM). This is the stored program concept (the Von Neumann model).
• New program? Load new instructions into memory. Nothing is rewired.
• Input devices send data in; output devices present results; secondary storage keeps programs when they are not running.

Inside the CPU

• Main memory is outside the CPU. The CPU must copy instructions and data from it before using them.

Buses

• A bus is a set of parallel wires carrying signals between the CPU, main memory and other components.
• The address bus carries memory addresses from the CPU to memory. It only goes one way.
• The data bus carries data and instructions between the CPU and memory, in both directions.
• The control bus carries control signals, such as "read" or "write" and clock pulses, to coordinate components.

The cycle that never stops

• Kept in step by the clock, the CPU repeats this cycle billions of times a second: the fetch–decode–execute (FDE) cycle.

• During fetch, the instruction's address goes out on the address bus and the instruction comes back on the data bus.

Embedded systems

• An embedded system is a computer built into a larger device to do one dedicated job.
• Examples: washing machine, microwave, thermostat, car engine management, traffic lights, vending machine.
• Usually small, cheap, low power and reliable. Its program is stored permanently, often in ROM.
• A general-purpose computer, such as a PC or laptop, runs many different programs.

Double the cores. Why isn't the computer twice as fast?

Clock speed

• The clock sends out regular electrical pulses that keep the CPU's parts in step.
• Clock speed is the number of clock cycles per second, measured in hertz (Hz).
• 1 GHz = 1,000,000,000 cycles per second, so 3.5 GHz is 3.5 billion.
• Higher clock speed: more fetch–decode–execute cycles each second, so more instructions are processed.

Cores

• A core is a complete processing unit. Each core can fetch, decode and execute instructions on its own.
• A dual-core CPU can process two instructions at the same time; a quad-core CPU, four.
• Doubling the cores rarely doubles the speed: some tasks cannot be split, and one core may wait for another's result.
• Software must be written to use several cores.

Cache

• Cache is a small amount of very fast memory on or very near the CPU.
• It holds frequently used instructions and data, saving slower trips to RAM.
• A larger cache holds more, so the CPU waits less. But cache is expensive, so it stays small.
• Cache has levels: L1 (smallest, fastest), then L2, then L3 (largest and slowest, but still faster than RAM).

Pull the plug and unsaved work is gone. So how does it boot?

Primary storage

• Primary storage (main memory) is memory the CPU can access directly.
• It holds the instructions and data the CPU is using now. Secondary storage is too slow for the CPU to work from directly.
• Main memory is made up of RAM and ROM.

RAM

• RAM (random access memory) holds the operating system, the programs that are open and the data they are using.
• The CPU can read and write RAM, so its contents change all the time.
• RAM is volatile: its contents are lost when the power is switched off. That is why unsaved work disappears.
• More RAM lets more programs and data be open at once without slowing down.

ROM

• ROM (read-only memory) is non-volatile: it keeps its contents with the power off.
• It can only be read. Its contents cannot normally be changed.
• ROM holds the start-up instructions (the BIOS or boot program). They load the operating system from secondary storage into RAM.

Registers and cache

• Registers are tiny memory locations inside the CPU holding the instruction, data or address in use right now.
• Cache is small, very fast memory on or near the CPU. It holds copies of frequently used instructions and data, so fewer slow trips to RAM are needed.
• Both are volatile. The faster the memory, the more it costs per byte, so fast memory is kept small.

Virtual memory

• When RAM is full, the operating system uses part of secondary storage as virtual memory.
• Data and programs not needed right now are moved out of RAM to virtual memory in blocks called pages, freeing space.
• When they are needed again, pages are swapped back into RAM.
• Secondary storage is much slower than RAM, so heavy swapping slows the computer down. Adding more RAM is the fix.

One storage firm runs 300,000+ spinning hard disks. Why not SSDs?

Why secondary storage?

• Secondary storage is non-volatile: it keeps the operating system, programs and files with the power off.
• It holds far more than RAM, for much less money per gigabyte.
• The CPU does not access it directly. Programs and data are copied into RAM (main memory) first.
• To choose a device, compare its capacity, speed, portability, durability, reliability and cost.

Magnetic storage

• A hard disk drive (HDD) stores data on spinning magnetic platters.
• Platters are divided into circular tracks, split into sectors.
• A moving read/write head uses an electromagnet to magnetise tiny areas (one way for 1, the other for 0) and to read them.
• Moving parts: slower, and easily damaged if dropped. But huge and cheap per gigabyte.

Solid-state storage

• Solid-state storage uses flash memory: electrical circuits that keep data with the power off. No moving parts.
• Examples: SSD (solid-state drive), SD card, USB flash drive.
• Fast, light, low power and robust. But it costs more per gigabyte, and cells wear out after many rewrites.

Optical storage

• A laser reads the surface of a spinning disc. Pits and lands reflect the light differently, giving 1s and 0s.
• To write, a laser changes the surface of a recordable disc.
• Examples: CD (about 700 MB), DVD (4.7 GB), Blu-ray (25 GB per layer).
• Cheap and light to post, but low capacity, slow and easily scratched. Downloads and flash drives have largely replaced it.

Cloud storage

• Cloud storage keeps your files on servers at a remote location, reached over the internet. The servers use magnetic and/or solid-state storage.
• Advantages: access from any device, anywhere; easy sharing; the provider handles backups and maintenance; buy more space when needed.
• Disadvantages: no internet, no access; slow connections mean slow transfers; ongoing fees; you rely on the provider to keep data secure and private.

Big ideas

• Every system can be described as input, process, output.
• Computational models can be sequential (one step at a time), parallel (parts at the same time) or multi-agent (many independent agents interacting).
• Simulation software runs a simplified model of the real world to test "what if?" safely and cheaply; it is only as good as its rules and data.
• Instructions and data are both stored in main memory; the CPU fetches, decodes and executes them, one after another, kept in step by the clock.
• The address bus carries addresses, the data bus carries data and instructions, the control bus carries control signals.
• An embedded system is a computer inside a larger device that does one dedicated job.
• Higher clock speed means more cycles, and so more instructions, every second.
• More cores help only when the work can be split; a larger cache means fewer slow trips to RAM.
• RAM is volatile and holds what is in use now; ROM is non-volatile and holds the start-up instructions.
• Registers and cache are small, fast, volatile memory close to the CPU.
• When RAM is full, virtual memory on secondary storage takes the overflow, but swapping is slow.
• Secondary storage is non-volatile and keeps data for the long term; choose it by capacity, speed, portability, durability, reliability and cost.
• Magnetic: huge and cheap per gigabyte, but moving parts. Solid state: fast and robust, but dearer per gigabyte.
• Optical discs are read by a laser (pits and lands): cheap but low capacity.
• Cloud storage is on remote servers: access anywhere, but only with an internet connection.