Question 4 of 6: Layered architecture and the OSI model

Question 4: Layered architecture and the OSI model (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.

Given. The OSI seven-layer reference model and a list of eight networking functions or protocols to be placed within it.

Find. The engineering advantage of layering, a one-sentence description of each of the seven layers, and the layer or layers at which each of the eight listed items belongs.

Part (a) — the advantage of a layered architecture. A layered architecture decomposes an intractably large design problem into a stack of smaller ones, each of which can be solved, tested and replaced on its own. Each layer offers a defined service to the layer above and consumes the service of the layer below through an interface that says what is provided but not how, so the internals of a layer are private. That single discipline buys several things at once. It buys interoperability, because two implementations that agree on the peer protocol and the service interface will work together even if they share no code — which is exactly why equipment from different vendors interconnects at all. It buys independent evolution: the physical layer under Ethernet has been rebuilt from thick coaxial cable to twisted pair to single-mode fibre, raising the rate by four orders of magnitude, without a single change to IP or TCP above it, and conversely IPv6 can replace IPv4 without touching the Ethernet MAC below. It buys reuse, since one transport implementation serves every application and one network layer serves every link technology — the classic hourglass, with IP at the waist. And it buys tractable fault-finding, because a fault can be localised by asking which layer’s guarantees are violated, which is how every practical diagnostic procedure from a cable tester to a protocol analyser is organised. The cost, which a complete answer should acknowledge, is some duplication of function between layers — error control appears at layers 1, 2 and 4 — and the loss of some cross-layer optimisation, which is why wireless designs frequently need information to leak across the layer boundary.

Part (b) — the seven layers. Figure 4.1 shows the stack; the descriptions are as follows. The physical layer (layer 1) transmits a raw bit stream over a medium, defining the voltages, modulation, carrier frequencies, bit timing, connectors and pin assignments. The data link layer (layer 2) turns a raw bit pipe into a frame-delivery service between two stations sharing one link, adding framing, physical addressing, error detection and, where the medium is shared, medium access control. The network layer (layer 3) delivers packets between any two hosts in an internetwork, choosing a route across intermediate nodes and providing global addressing, fragmentation and congestion handling. The transport layer (layer 4) provides end-to-end delivery between processes on the two end systems, optionally adding reliability, sequencing, flow control and end-to-end congestion control that the intervening network does not offer. The session layer (layer 5) organises and synchronises the dialogue between two applications, managing who may transmit, checkpoints and recovery of a long interaction. The presentation layer (layer 6) deals with the syntax and semantics of the information transferred, converting between local data representations and a common transfer syntax, and performing compression and encryption. The application layer (layer 7) contains the protocols the user’s program itself speaks, such as HTTP, SMTP, FTP and DNS.

Part (c) — placing the eight items. The mapping is collected in the table below, and four of the eight deserve a comment because they legitimately span more than one layer. Ethernet as standardised in IEEE 802.3 is both a data link specification (the MAC sublayer: framing, 48-bit addressing, CSMA/CD) and a family of physical layer specifications (10BASE-T, 100BASE-TX, 1000BASE-LX), so a complete answer names layers 1 and 2. Error-control coding appears twice for genuinely different reasons: forward error correction and line coding sit in the physical layer, where redundancy is added to combat the raw bit error rate, while the frame check sequence and any retransmission protocol sit in the data link layer, and the transport layer adds an independent end-to-end checksum because the layers below protect only individual hops. The TCP/IP protocols are deliberately plural — TCP is a transport protocol at layer 4 and IP is a network protocol at layer 3 — and this is the item that most often loses marks when answered with a single layer. Finally, confirmation of end-to-end delivery across multiple hops is the transport layer and not the data link layer, precisely because a link-layer acknowledgement confirms only the hop just crossed; only an end-to-end acknowledgement proves that the far host received the data, which is the argument of the end-to-end principle.