22-Elec-B4 Information Technology Networks · May 2014
Nivaar worked solution (AI-drafted; not reviewed by a licensed engineer)
Paper format. Professional Engineers of Ontario annual examination, 07-Elec-B4 Information Technology Networks, May 2014. Three hours, closed book, one PEO-approved non-programmable calculator permitted. Marks are printed in the left margin; the cover page states that there are five questions and that any four constitute a complete paper worth 100 marks. All five questions and every sub-part are answered below, because this set is intended as a study resource rather than as a sat examination.
Reference texts. A. Leon-Garcia and I. Widjaja, Communication Networks: Fundamental Concepts and Key Architectures, 2nd ed. — the text listed by the Engineers Canada syllabus for this examination code; J. F. Kurose and K. W. Ross, Computer Networking: A Top-Down Approach, 8th ed.; A. S. Tanenbaum and D. J. Wetherall, Computer Networks, 5th ed.; W. Stallings, Wireless Communications and Networking, 2nd ed.; T. S. Rappaport, Wireless Communications: Principles and Practice, 2nd ed. Normative documents cited: ISO/IEC 7498-1 (the OSI reference model), IEEE 802.3 (CSMA/CD), IEEE 802.11 (wireless LAN), 3GPP TS 23.401 (the LTE Evolved Packet Core), RFC 768 (UDP) and RFC 959 (FTP).
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.
The distinction is not about the physical plant — both disciplines run over the same fibre, copper and radio — but about when and for how long network capacity is committed to a conversation.
In a circuit-switched network, a signalling phase precedes any data transfer. The signalling messages walk the path from source to destination and cause every switch along it to reserve a share of each outgoing link — a wavelength, a frequency band, or a recurring time slot — and to write a cross-connect entry that welds the inbound share to the outbound share. That capacity belongs to the connection until it is torn down, whether or not anything is being sent through it. Once the circuit exists, switching is trivial: there is no address to look up and no queue to join, so the delay is just propagation plus a fixed, tiny switching latency, and it does not vary. Congestion appears only at setup, as blocking: if no free share exists on some link, the call is refused outright rather than degraded.
In a packet-switched network there is no reservation and no setup. The message is broken into packets, each carrying a header with the full destination address, and each packet is handled independently: a switch receives it in full, reads the header, looks the destination up in a forwarding table, queues the packet on the chosen outgoing link and transmits it when the link is free. This is store-and-forward operation, and the queue is where the statistical multiplexing gain lives — idle moments in one conversation are immediately usable by another. The costs are the mirror image of the benefits: queueing delay that varies with load (jitter), packet loss when a buffer fills, possible reordering when successive packets take different routes, and a header carried on every packet.
| Property | Circuit switching | Packet switching |
|---|---|---|
| Capacity commitment | Reserved at setup, held until teardown | None; shared statistically, packet by packet |
| Setup phase | Required (signalling) | None (connectionless forwarding) |
| Per-unit overhead | Address sent once, at setup | Full header on every packet |
| Delay | Constant, minimal, no jitter | Variable: transmission + queueing at every hop |
| Behaviour under overload | Blocking — new calls refused, existing calls unaffected | Delay grows, then loss; all flows degrade together |
| Utilisation on bursty sources | Poor (capacity idles) | High (idle capacity is reused) |
| Response to a link failure | Call drops; must be re-established | Routing re-converges; the flow continues on a new path |
GSM was designed as a digital telephone network that happened to be wireless, and its architecture is circuit-switched end to end. A voice call occupies one recurring time slot on the air interface for its entire duration and a 64 kbit/s circuit through the Mobile Switching Centre in the core, held whether the subscriber is speaking or silent. Packet data was retro-fitted much later, as GPRS and then EDGE, and it was fitted alongside the circuit domain rather than replacing it: a GSM/UMTS operator ran two parallel core networks, a circuit-switched domain (MSC, visitor and home location registers) for voice and SMS, and a packet-switched domain (SGSN, GGSN) for data, with the radio resource divided between them.
LTE removes the circuit-switched domain entirely. The Evolved Packet Core has no MSC and no circuits; every service, voice included, is carried in IP packets over an EPS bearer, and the eNodeB scheduler allocates resource blocks to users afresh every millisecond. Voice becomes an application: VoLTE runs the call through the IP Multimedia Subsystem over a dedicated bearer with a guaranteed bit rate and the highest scheduling priority class, while early deployments without IMS simply handed the subscriber back to the legacy 2G/3G circuit domain for the duration of a call (Circuit-Switched Fallback).
The engineering consequences are worth stating, because they are what the question is really asking about. Voice and data now share one pool of radio resource, so silence in a conversation is immediately available to somebody else's download — a real capacity gain, since natural speech is active barely half the time. The operator maintains one core network instead of two. Against that, the quality guarantee that a reserved circuit provided for free must now be engineered: it depends on the scheduler honouring the QoS class, on admission control, and on the packet delay budget being met hop by hop. A circuit either exists at full quality or does not exist at all, whereas a packet bearer can degrade gradually, which is a different operational problem.
| Method | Most useful for | One-sentence reason |
|---|---|---|
| i. FDMA | Circuit switching | Each user is given an exclusive frequency band for the duration of the call, which is precisely a reserved circuit realised in the frequency domain. |
| ii. TCP/IP | Packet switching | IP forwards independently addressed datagrams hop by hop with no reservation, and TCP exists exactly because that underlying service is unreliable, unordered and variably delayed. |
| iii. TDMA | Circuit switching | A user is assigned the same time slot in every frame for the whole call, which is a reserved circuit realised in the time domain rather than the frequency domain. |
| iv. Ethernet | Packet switching | Every frame carries its own destination MAC address and is forwarded on its own merits, with no capacity reserved for any talker. |
Two qualifications earn marks here. First, FDMA and TDMA are multiple-access schemes rather than switching disciplines, and a packet system can be built on top of either — GPRS assigns GSM time slots dynamically, packet by packet, and 802.11 uses one frequency channel for contention-based packet access. What makes them circuit-like is the classical use: a fixed, recurring allocation held for the call. Second, Ethernet's original CSMA/CD form is the opposite of reservation in the strongest sense, since stations contend for the medium and back off after collisions; switched Ethernet keeps the connectionless frame-forwarding model while removing the contention.
Packet switching is decisively better, and the reason is statistical multiplexing. A bursty source has a peak rate much higher than its average rate. Circuit switching must reserve the peak, because the circuit is fixed for the duration, so the reserved capacity idles whenever the source is quiet. Packet switching commits capacity only while a packet is actually being transmitted, so the troughs of one source are filled by the peaks of others.
A concrete illustration with a 1 Mbit/s link and sources that peak at 100 kbit/s and are active 10 % of the time: circuit switching admits
$$ n_{\text{circuit}} = \frac{R_{\text{link}}}{R_{\text{src}}} = \frac{1\ \text{Mbit/s}}{100\ \text{kbit/s}} = \boxed{10\ \text{sources}} $$
and those ten sources use, on average, only 10 % of the link. Under packet switching, admit 35 such sources instead. The mean number simultaneously active is $35 \times 0.10 = 3.5$, and the link overflows only when more than ten are active at once, whose binomial probability is
$$ P(X > 10) = \sum_{k=11}^{35}\binom{35}{k}(0.1)^{k}(0.9)^{35-k} = 4.2\times10^{-4} $$
— about four occasions in ten thousand, and even then the excess is absorbed by the buffer rather than lost. The same link therefore carries three and a half times as many sources at a negligible quality penalty. The burstier the source — that is, the higher its peak-to-mean ratio — the larger this gain becomes, which is why interactive and web traffic, whose peak-to-mean ratio can exceed 100, is never carried on reserved circuits.
Circuit switching is better by construction, although in practice the answer is more interesting than that. A circuit has no queue anywhere along its path, so its delay is the propagation time plus a fixed switching latency: it is constant, it is known in advance, and it does not depend on what anybody else on the network is doing. Nothing has to be engineered to make that true. Packets, by contrast, queue at every hop behind other people's traffic, so the delay is a random variable whose tail grows sharply as link utilisation approaches unity, packets may be reordered, and a lost packet costs a retransmission timeout.
The qualification is that modern networks meet hard delay constraints with packet switching anyway, because the efficiency gain from part (d) is too large to give up. They do it by adding machinery that a circuit provides for free: classification and priority scheduling so that delay-sensitive packets are served first, admission control and traffic shaping so that the high-priority class stays below the load at which its queue grows, deliberate over-provisioning of the backbone, and a playout buffer at the receiver that converts variable delay into a constant, slightly larger delay. VoLTE is precisely this: a packet bearer engineered to a 100 ms packet delay budget so that it behaves like the circuit it replaced. If the requirement is a guarantee rather than a well-engineered expectation, the circuit still wins.
