Delays in packet switched network

In a packet-switched network a packet passes through many nodes (routers) from source to destination. At each node it suffers a nodal delay:

d_nodal = d_proc + d_queue + d_trans + d_prop

 Host A ---[ Router ]------------- Host B
            ^   ^   ^       ^
         proc queue trans  prop

1. Processing delay (d_proc): the time the router takes to examine the packet header, check for bit errors and decide the output link (routing table lookup). It is usually microseconds or less.

2. Queuing delay (d_queue): time the packet waits in the output buffer before it can be sent because other packets are ahead. It depends on traffic intensity La/R (L = packet size in bits, a = average arrival rate in packets/s, R = link rate in bps). If La/R is near 1 the delay becomes very large; if La/R > 1, the buffer fills and packets are lost. It varies from packet to packet (jitter).

3. Transmission delay (d_trans): time to push all the bits of the packet onto the link. d_trans = L / R Example: L = 1500 bytes = 12,000 bits, R = 10 Mbps, so d_trans = 12,000 / 10^7 = 1.2 ms.

4. Propagation delay (d_prop): time for a bit to travel along the link from one end to the other. d_prop = d / s, where d is link length and s is propagation speed (about 2 x 10^8 m/s in fibre/copper). Example: 2,000 km link, so d_prop = 2 x 10^6 / 2 x 10^8 = 10 ms.

Note: transmission delay depends on packet size and link rate; propagation delay depends on distance and medium. They are different.

End-to-end delay for N links in a path with no queuing: roughly N(d_proc + d_trans + d_prop). In a store-and-forward router the whole packet must arrive before it is forwarded, so d_trans appears at each hop.

Ways to reduce delay: faster links (lower d_trans), fast hardware (lower d_proc), proper queue management and QoS (lower d_queue), shorter routes or CDNs (lower d_prop).