Question 5 of 5: Layered architecture

Question 5: Layered architecture (25 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

(a) The advantage of a layered architecture

Layering decomposes an intractable problem — move arbitrary data between arbitrary machines over arbitrary media — into a stack of small problems, each solved once. Each layer offers a defined service to the layer above through an interface, and implements that service by using the layer below and by talking to its own peer at the far end through a protocol. Because the interface is what the neighbour depends on, and the protocol is private to the layer, an implementation may be replaced entirely without disturbing anything else: swapping copper for fibre changes only layer 1, and moving an application from Ethernet to a wireless link changes nothing above layer 2. That property — modularity with substitutability — is the advantage the question is asking for, and everything else follows from it.

The consequences are practical rather than abstract. Complexity is bounded, so a designer can reason about one layer at a time. Standardisation becomes possible, so equipment from different vendors interworks as long as each honours the interface, which is precisely what let the Internet grow without a central authority. Interoperability and reuse follow: one transport protocol serves every application, and one application runs over every medium, so $m$ applications and $n$ media need $m+n$ pieces of work rather than $m \times n$. Development and testing can proceed in parallel, and a fault can be localised to a layer instead of hunted across the whole system. The cost, worth acknowledging, is some duplication of function between layers and the processing overhead of headers and copies at each boundary — which is why real stacks such as TCP/IP collapse the OSI seven into four or five.

(b) The seven layers of the OSI model

1. Physical. Transmits raw bits over a physical medium, defining the mechanical, electrical and timing characteristics — connectors, voltage levels, modulation and detection, bit rate — so that a bit sent as a one arrives as a one.
2. Data link. Turns the raw bit pipe into a link that appears free of undetected errors between two directly connected nodes, by framing the bits, adding error detection, controlling flow across the link, and — in its medium access control sublayer — arbitrating access when the link is shared.
3. Network. Delivers packets from a source host to a destination host across an arbitrary sequence of intermediate networks, which requires global addressing, routing, fragmentation to fit each link, and the network-wide part of congestion control.
4. Transport. Provides end-to-end delivery between application processes on the two hosts, multiplexing them by port number and, if the service is reliable, adding sequencing, acknowledgement, retransmission, flow control and congestion control.
5. Session. Establishes, manages and terminates a dialogue between two applications, adding dialogue control, token management for critical operations and checkpointing so a long transfer can resume after a failure rather than restart.
6. Presentation. Gives meaning to the bits by handling the syntax and semantics of the information transferred — character-code and data-representation conversion, abstract syntax notation, compression and encryption — so that machines with different internal formats understand one another.
7. Application. Provides the network services the user’s program actually invokes, such as electronic mail, file transfer, directory lookup and the web, and is the only layer with no layer above it to serve.

(c) Where each item is used or found

Each item is placed by asking what job it does, not which software product happens to contain it. Two of the seven span more than one layer, and saying so is part of the answer.