Layered architecture: 22-Elec-B4 Information Technology Networks, May 2015, Q4

Question 4: 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.

Part (a) — the advantage of layering. Layering is the application of modular decomposition to a system too large to design as one piece. Each layer offers a defined service to the layer above through a defined interface, and implements that service by using the service of the layer below plus a protocol spoken with its peer at the far end. Because the interface is all that the layer above may depend on, the implementation behind it can be replaced freely.

That single property buys several concrete advantages. Implementations can be changed independently: replacing copper with fibre, or Ethernet with Wi-Fi, changes layers 1 and 2 and nothing else, so no application is rewritten. Work can be divided among different organisations and standardised separately, which is why the IEEE can own Ethernet, the IETF can own IP and TCP, and the W3C can own the web, with no committee needing to understand all three. Interoperability follows from the same discipline, since two products interwork if they agree on the protocol at each layer, regardless of internal design. Faults are localised, because a problem can be bisected layer by layer — if the link carries frames but the application fails, the fault lies above layer 2. And a layer is reused rather than reinvented: IP runs over every link technology and every application runs over IP, the "hourglass" shape that let the Internet absorb technologies nobody had thought of when it was designed. The cost, worth a sentence for a complete answer, is header overhead, some duplication of function (error detection appears at layers 2 and 4), and occasional performance loss where hiding information is exactly wrong — classically, TCP interpreting a wireless bit error as congestion.

Part (b) — the seven layers.

Figure 4.1 — The OSI reference model. Layers 1 to 3 are implemented hop by hop in the subnet as well as in the end systems; layers 4 to 7 exist only in the end systems.

No.	Layer	One-sentence description
7	Application	Provides network services directly to user processes and defines the messages they exchange, such as HTTP, SMTP, FTP and DNS.
6	Presentation	Resolves differences in the representation of data between the two end systems — character encoding, data-structure marshalling, compression and encryption — so that the application layer can exchange meaning rather than bytes.
5	Session	Establishes, manages and synchronises a dialogue between two applications, including token management for who may transmit and checkpointing so a long transfer can resume after a failure.
4	Transport	Provides end-to-end transfer of data between processes on the two end systems, multiplexing them by port and optionally adding reliability, ordering, flow control and congestion control.
3	Network	Delivers packets from the source host to the destination host across a multi-hop subnet, which requires global addressing, routing and per-hop forwarding.
2	Data link	Turns the raw bit pipe into a reliable frame-by-frame link between two directly connected nodes, adding framing, error detection, and — on a shared medium — medium access control.
1	Physical	Transmits raw bits over the medium, defining voltages, timing, modulation, connectors and pin assignments.

A useful sanity check on the split: layers 1 to 3 are present in every intermediate node of the network, whereas layers 4 to 7 exist only in the two end systems — a router forwards packets and therefore needs a network layer, but it has no business terminating a TCP connection.

Part (c) — locating each item in the model.

Item	OSI layer(s)	Justification
i. The HTTP protocol	7 — Application	It defines the request and response messages exchanged between a browser process and a server process, which is precisely the application layer's job.
ii. The TCP/IP protocol	4 and 3 — Transport and Network	The name is a pair, not one protocol: TCP is the transport-layer protocol providing a reliable process-to-process byte stream, while IP is the network-layer protocol providing host-to-host packet delivery.
iii. Ethernet	2 and 1 — Data link and Physical	IEEE 802.3 specifies both the MAC sublayer of the data link layer (framing, addressing, CRC, CSMA/CD) and the physical layer (10BASE-T, 100BASE-TX and the rest).
iv. Routing	3 — Network	Choosing a path across a multi-hop subnet, and populating the forwarding tables that implement it, is the defining function of the network layer.
v. End-to-end packet delivery	4 — Transport	Only the transport layer is truly end-to-end in the OSI sense, since it exists solely in the two end systems and delivers to a process; the network layer is host-to-host and is re-executed at every intermediate node.
vi. Encryption and decryption	6 — Presentation	Transforming the representation of data for confidentiality is a presentation function in the OSI model, though in the TCP/IP world the same job is done by TLS between layers 4 and 7, and link encryption exists at layer 2.
vii. Character display, such as ASCII	6 — Presentation	Agreeing a character encoding so that the same byte means the same character at both ends is the classic presentation-layer syntax translation.
viii. Signal voltages	1 — Physical	Voltage levels, together with timing and connectors, are what the physical layer specifies in order to put a bit on the wire.

Topic. Protocol layering, the OSI reference model, and the service / interface / protocol distinction.

Key relations. Layer $n$ provides a service to layer $n+1$ and consumes the service of layer $n-1$; peer entities at layer $n$ communicate by a layer-$n$ protocol. Encapsulation: a layer-$n$ PDU becomes the payload (SDU) of layer $n-1$, so the frame on the wire is $H_2 \,|\, H_3 \,|\, H_4 \,|\, \text{data} \,|\, T_2$. Mnemonic bottom-up: Please Do Not Throw Sausage Pizza Away.

Why this works. The value of layering comes from the interface, not from the number of layers. Because layer $n+1$ may only invoke the service primitives layer $n$ publishes, it cannot come to depend on how that service is achieved, and so the implementation can change without breaking anything above it. That is why the Internet outlived every transmission technology it started on. The OSI model itself was never deployed as a protocol suite — TCP/IP won — but it survives as the vocabulary the industry argues in, and its seven layers remain the reference against which real stacks are described, even where they map imperfectly (TLS and the whole session layer being the obvious cases).

Common pitfall. Answering "TCP/IP" with a single layer number; the item names two protocols at two layers and the marks are for spotting that. Second, putting routing at layer 2 because switches "route" — a switch forwards frames using MAC addresses within one link-layer network, which is layer 2 bridging, while routing between networks using global addresses is layer 3. Third, describing the presentation layer as "formatting for display", which confuses the syntax of the data on the wire with what an application chooses to draw on a screen.

Source. ISO/IEC 7498-1, the OSI Basic Reference Model; Tanenbaum and Wetherall, Computer Networks, 5th ed., §1.4.1; Leon-Garcia and Widjaja, Communication Networks, 2nd ed., §2.2 (layered architectures) and §2.3.

Question 4 of 5: Layered architecture

Question 4: Layered architecture (25 marks)