Switching and performance metrics

Circuit switching

Definition. Circuit switching establishes a reserved end-to-end path and allocates resources for the duration of a session. After setup, data can flow with predictable capacity and little per-packet routing overhead. Reserved capacity can be wasted during silence. Example. A 64 kbit/s circuit reserved for a call remains allocated even when the speaker pauses. Exam point. Circuit switching has setup delay, then dedicated resource allocation.

Packet switching

Definition. Packet switching divides data into packets that share link capacity with other traffic. Statistical multiplexing improves utilization but can create queueing delay and loss under congestion. Each packet may be forwarded independently or within a logical connection. Example. An IP router forwards datagrams from many users over one outgoing link. Exam point. Shared bandwidth does not mean every flow receives a fixed rate.

Datagram and virtual-circuit networks

Definition. In a datagram network, each packet carries a destination address and routers make forwarding decisions independently. A virtual-circuit network establishes logical forwarding state before data transfer; packets carry a short virtual-circuit identifier that is locally meaningful. The logical path is not necessarily a physically dedicated circuit. Example. IP is a datagram service; a virtual-circuit network can map identifier 12 to identifier 41 at a switch. Exam point. Virtual-circuit identifiers can change at each hop.

Feature	Datagram	Virtual circuit
Setup	not required per flow	path state established first
Packet header	destination address	local VC label/identifier
Forwarding	per-packet lookup	connection table lookup
Route changes	packets can take different paths	usually follow established VC
Failure handling	later packets may be rerouted	VC state/path may need repair

Statistical multiplexing

Definition. Statistical multiplexing lets sources share a link according to their instantaneous demand. It is efficient when sources are bursty and not all peak simultaneously. When aggregate arrival rate exceeds service rate, queues grow and packets may be delayed or dropped. Example. Ten users each needing 1 Mbit/s only 10% of the time can share a 10 Mbit/s link in favorable conditions, though peak guarantees are not provided. Exam point. Average load below capacity does not eliminate short-term congestion.

Store-and-forward packet delay

Definition. A store-and-forward switch must receive a whole packet before transmitting it on the next link. For a packet of L bits crossing N links of rate R with negligible processing, queueing, and propagation, transmission delay is $NL/R$. If rates differ, sum $L/R_i$ over links. Example. A 12000-bit packet crosses three 4 Mbit/s links: total transmission time is $3(12000/4\times10^6)=9$ ms. Exam point. Store-and-forward transmission delay accumulates at each hop.

Pipelined packet transmission

Definition. On equal-rate links, after the first packet begins leaving the first link, different bits of the packet can occupy different links. For a single packet, the first bit still needs one transmission time per hop; for a stream of packets, completion spacing can be one packet transmission time after the pipeline fills. Example. For two 1000-bit packets on three 1 Mbit/s links, with no propagation or processing, total completion time is $(N+P-1)L/R=(3+2-1)8$ ms = 32 ms. Exam point. Do not multiply every packet by the number of links when transmissions can pipeline.

Worked end-to-end delay with propagation

Definition. One-way end-to-end delay can be calculated by summing transmission, propagation, processing, and queueing at each relevant stage. If packet lengths differ by link because of headers or fragmentation, use the length actually transmitted on that link. The problem’s switching model determines whether links overlap. Example. A 10000-bit packet crosses two 2 Mbit/s links, each with 5 ms propagation, using store-and-forward and zero queue/processing. Exam point. Transmission and propagation both occur on each hop, but propagation is not multiplied by packet length.

Step-by-step:

1. Transmission per link is $10000/(2\times10^6)=0.005$ s = 5 ms.
2. Propagation per link is 5 ms.
3. Store-and-forward requires reception before sending on link 2.
4. Total is 2(5+5)=20 ms.