Networks and Web Technologies

The application layer is where the protocols used directly by software operate, such as HTTP for web pages and SMTP for sending mail, and it is responsible for formatting the data appropriately for the service being used. The transport layer takes that data and divides it into segments, adding port numbers so the receiving machine knows which application the data belongs to, and where TCP is used it also numbers the segments and manages acknowledgements and retransmission to guarantee complete, correctly ordered delivery. The network layer, sometimes called the internet layer, attaches the source and destination IP addresses and is responsible for routing each packet across the intervening networks towards its destination. The link layer handles the actual transmission over the physical medium, adding MAC addresses so that frames reach the correct device on the local network, and it is where Ethernet and Wi-Fi operate.

UDP is connectionless: it sends datagrams without establishing a connection, without acknowledgements and without retransmitting anything that is lost, so it carries far less overhead than TCP and delivers data with noticeably lower latency. That matters enormously in a live call, where the total delay between speaking and being heard determines whether a conversation is usable at all. The decisive point is what would happen if TCP were used instead. If a packet carrying a fiftieth of a second of audio were lost, TCP would detect the gap and retransmit it, but by the time the replacement arrived that moment in the conversation would already have passed, so the retransmitted data is worthless while the stall waiting for it is very noticeable. With UDP the missing fragment simply produces a momentary glitch that the human ear largely ignores, and the call continues smoothly. In short, for live media a small amount of loss is far preferable to delay, which is precisely the trade UDP makes.

Client side processing is code, usually JavaScript, that is downloaded and executed by the user's own browser, whereas server side processing runs on the web server and only the results are sent to the client. The advantage of client side processing is responsiveness: a form field can be checked the moment the user leaves it, with no request to the server and no waiting, which makes the interface feel immediate and reduces load on the server. The critical weakness is that anything running on the user's machine is entirely under their control. They can read the source, disable JavaScript altogether, modify the code with browser developer tools, or bypass the page completely and send a crafted request directly to the server. Client side validation therefore provides usability but no security whatsoever. Server side validation is authoritative because the user cannot alter or bypass it, and it is the only check that can be trusted to protect the database from invalid or malicious data. Both are used together: client side for a good experience, server side because it is the only one that actually enforces anything.

PageRank estimates the importance of a page from the structure of links pointing at it, treating each incoming link as a vote of confidence. Crucially the votes are not counted equally: a link from a page that is itself highly ranked contributes far more than a link from an obscure one, so the rank of a page depends recursively on the ranks of the pages linking to it. Each page's voting power is also divided among all the pages it links out to, so a page with a thousand outgoing links passes only a thousandth of its rank to each, which prevents importance being manufactured simply by linking to everything. A damping factor, conventionally 0.85, is applied to model the probability that a person following links eventually stops and starts somewhere new, and it also guarantees that the mathematics converges rather than concentrating all rank in a small group of pages. Because each page's rank depends on others which are themselves being calculated, the algorithm is iterative: every page begins with an equal rank, the formula is applied to every page, and this repeats until the values stop changing significantly.

In a client server network, dedicated server machines provide resources and services which client machines request, so files, authentication and applications are centralised. In a peer to peer network every machine is an equal that acts as both client and server, sharing its own resources directly with the others. The client server model's strength is control: data can be backed up centrally in one place, user accounts and permissions are administered from a single point, and software updates and security policies are applied consistently across the whole network, which is essential once an organisation grows beyond a handful of machines. Its weaknesses are cost and dependency, since server hardware and specialist staff are expensive, the server represents a single point of failure whose loss affects every user, and it can become a bottleneck if demand exceeds what it can serve. Peer to peer inverts this. It is cheap to set up because there is no server to buy, load is naturally distributed across the participants so it scales well as more peers join, and there is no single point of failure. In exchange it offers no central backup, so data protection depends on each user individually, no central management of security or updates, and availability depends entirely on which peers happen to be switched on, so a file may simply be unreachable. In practice client server suits organisations needing control and reliability, while peer to peer suits small informal networks and distributed content sharing.