Hardware and communication

Every computer system, from a phone to a server, has the same main parts:

They are connected by buses, sets of parallel wires that carry signals between components:

• Data bus: carries data and instructions. It works in both directions.
• Address bus: carries the memory address to read from or write to. One direction only, from the processor. Its width sets how much memory can be addressed: a 32-bit address bus can address 2³² locations.
• Control bus: carries control signals such as read, write, clock and interrupt requests.

External devices connect through ports and controllers, for example USB, HDMI, Thunderbolt or a network port, or wirelessly.

In the Von Neumann architecture, programs and data are stored together in the same main memory (the stored program concept). The processor fetches instructions one at a time and executes them.

The processor contains:

• Arithmetic logic unit (ALU): does calculations and logical comparisons.
• Control unit (CU): decodes instructions and sends control signals to coordinate everything else.
• Clock: sends regular pulses that keep components in step.
• Registers: very fast storage locations inside the processor.

The Von Neumann bottleneck: instructions and data share one memory and one set of buses, so they cannot be fetched at the same time. The processor can be left waiting for memory.

Harvard architecture keeps instructions and data in separate memories with separate buses, so both can be fetched at once. It is common in microcontrollers and digital signal processors. Most modern processors are a mix: one main memory, but separate caches for instructions and data close to the processor.

• SRAM (static RAM) is fast but expensive and takes more space per bit, so it is used for cache.
• DRAM (dynamic RAM) must be refreshed constantly but is cheaper and denser, so it is used for main memory.

Cache is small, fast memory between the processor and main memory. When the processor needs something, it checks the cache first:

• Cache hit: it's there, so it arrives quickly.
• Cache miss: it has to come from slower main memory, and a copy is kept in the cache.

Caching works because programs tend to reuse the same data and instructions (loops) and use items stored next to each other (arrays, sequential code).

Most processors have several levels: L1 (smallest, fastest, usually per core), L2, and L3 (largest, slowest, often shared between cores).

The memory hierarchy: moving away from the processor, memory gets larger, cheaper per byte and slower: registers → cache → RAM → secondary storage.

The same idea is used elsewhere: a disk cache keeps recently used disk data in RAM, and a web browser cache keeps copies of pages and images on your device.

Parallel processing means carrying out more than one instruction, or more than one part of a task, at the same time.

• Multi-core processors: several processing cores on one chip, each able to run its own instructions.
• Multiprocessor systems: several separate processors, as in servers and supercomputers.
• SIMD (single instruction, multiple data): one instruction is applied to many data items at once. Graphics processors (GPUs) work like this, which suits graphics, video and machine learning.
• MIMD (multiple instruction, multiple data): different processors run different instructions on different data. Multi-core processors are MIMD.
• Pipelining: while one instruction is being executed, the next is decoded and the one after that fetched, so the stages overlap.

Advantages: tasks that can be split up finish faster, more programs can run at once, and work can be shared to keep each core cooler and more efficient.

Disadvantages: programs must be written to split the work, some tasks cannot be split, and coordinating the processors adds overhead. Unit 4 looks at these limits in detail.

The processor repeats this cycle for every instruction. Square brackets mean "the contents of".

Fetch

1. MAR ← [PC] (the address of the next instruction is copied to the MAR)
2. PC ← [PC] + 1 (the program counter moves on to the next instruction)
3. MDR ← [Memory] addressed by the MAR (the instruction travels along the data bus)
4. CIR ← [MDR] (the instruction is copied into the current instruction register)

Decode: the control unit works out what the instruction means: its operation code (opcode) and operand.

Execute: the instruction is carried out.

Reading data from RAM into a register. If the instruction is LDA 20 (load the value stored at address 20 into the accumulator), the execute stage is:

1. MAR ← 20
2. MDR ← [Memory] at address 20
3. ACC ← [MDR]

After executing, the processor checks for interrupts before starting the next fetch.

Choose a device by thinking about the user, the data and the environment.

In an exam, always link the device to the situation: say why it suits that user or environment.

Functional characteristics to compare: capacity, access speed, transfer rate, cost, portability, durability and reliability.

A file is stored in blocks on a disk. When files are deleted, gaps are left. A new file may not fit in one gap, so it is split across several gaps in different places: the file is fragmented.

Consequences on a hard disk: to read a fragmented file the read/write head has to move to several places, so loading takes longer. The disk does more work and the free space is broken into small pieces.

Defragmentation reorganises the disk so each file's blocks are next to each other (contiguous) and the free space is gathered together. Files load faster afterwards.

SSDs should not be defragmented. They have no moving head, so access time is the same wherever the data is, and each extra write wears the flash cells. Instead the operating system uses the TRIM command to manage free space efficiently.

• LAN (local area network): covers one site, such as a school or office. The organisation owns the hardware.
• WAN (wide area network): covers a wide area, using infrastructure such as leased lines owned by telecoms companies. The internet is the largest WAN.

Client–server: a central server provides services (files, email, log-in) to client computers. Easier to manage, secure and back up, but the server is a single point of failure.

Peer-to-peer: every computer is equal and shares its resources directly. Cheap and simple, but harder to manage and secure.

Devices communicate by splitting data into packets. Each packet carries addressing information, so it can be delivered and reassembled. Data transmission covers packets and routing in detail.

A standard is an agreed specification that many manufacturers follow, set by bodies such as the IEEE or IETF. A protocol is a set of rules for communication: the format of messages, how they are sent and how errors are handled.

Why they matter: devices and software from different manufacturers can work together, buyers aren't locked into one supplier, and products are cheaper and easier to develop.

TCP/IP is a stack of four layers, each with its own protocols: application (HTTP, SMTP), transport (TCP, UDP), internet (IP) and link (Ethernet, Wi-Fi). Each layer only deals with the layer above and below, so one layer can change without affecting the others.

Handshaking is the exchange of signals between two devices before data is sent, to check that both are ready and to agree the rules for the transfer, such as the speed, the protocol and the error checking.

TCP's three-way handshake sets up a connection:

1. SYN: the client asks to connect.
2. SYN-ACK: the server agrees and acknowledges.
3. ACK: the client acknowledges, and data can flow.

Handshaking is also used between a computer and a device such as a printer: the printer signals when it is ready, busy or out of paper, so data is not sent faster than it can be handled.

The internet is a worldwide network of networks. The World Wide Web is just one service that runs on it, alongside email, file transfer, streaming and more.

• Internet service providers (ISPs) connect homes and businesses.
• Backbone links: high-capacity fibre connections between major networks and countries, including undersea cables.
• Internet exchange points (IXPs): places where networks connect and swap traffic.
• Routers forward packets along the best available route; if a link fails, traffic goes another way.
• DNS servers turn names such as helloworld.org.uk into IP addresses.
• Data centres host servers for websites and cloud services.

No single organisation owns the internet. It works because every part follows the same protocols (TCP/IP).