Delay, Loss & Throughput

Question 1

Delay Overview

Answer 1

• Delay is a measure of the time taken for a packet to travel across the network
• Measured in fractions of a second
  
  • Usually milliseconds – i.e., thousandths of a second

Question 2

Types of Delay

Answer 2

• There are four main types of delay:
  
  • Processing Delay
  • Queuing Delay
  • Transmission Delay
  • Propagation Delay

Question 3

Processing Delay

Answer 3

• Time taken for a device to examine a packet’s header and decide where to direct the packet
• May include a check of bit-level errors (as caused during
  transmission)
• Typically, this delay is very small
  
  • Usually microseconds – i.e., millionths of a second
• Can vary depending on how busy the device is

Question 4

Queuing Delay

Answer 4

• Once a packet has been processed, it will join a queue
  – It waits here to leave the device
  – It will not be sent until it reaches the head of the queue
• Queuing Delay is the time spent waiting in the queue before a packet is transmitted
• The length of the queue, and thus the delay, is dependent on the congestion level of the node/router
• N.B. A queue only develops if the packet arrival rate to link (temporarily) exceeds output link capacity

Question 5

Transmission Delay

Answer 5

• Packets are transmitted once they reach the head of the queue…
• Networks are Store-and-Forward
  
  • An entire packet must be received before it’s forwarded
• Transmission Delay is the amount of time required to push (transmit) all of the packet’s bits onto the link
• Transmission Delay = L/R
• L: packet length (bits)
• R: link bandwidth (bps)

Question 6

Propagation Delay

Answer 6

• Once a bit has been pushed onto a link, it needs to propagate to the next device
• Each link has an associated Propagation Delay
  
  • Measured as the time needed to get over the link from one end to another
  • Propagation delay is dependent on the physical type of the link*
• for wireless communications – a little less than the speed of light
• Propagation Delay = d/s
• d: length of physical link
• s: link’s propagation speed (~2x108 m/sec)

Question 7

Transmission vs Propagation Delay

Answer 7

• A subtle difference, but important!
• Transmission Delay is the time required to push a packet out
  – It’s a function of the packet’s length and the transmission rate of the link interface
  – Nothing to do with the distance between the two devices
• Propagation Delay is the time taken for a bit to propagate from one device to the next
  – Function of the link technology and the distance between two devices
  – Nothing to do with the packet’s length or the transmission rate of the link interface

Question 8

Convoy Analogy

Answer 8

• Car ~ bit; convoy ~ packet
• Cars “propagate” at 100 km/hr
• Toll booth takes 12 sec to service
  a car (bit transmission time)
• Q: How long until convoy is lined
  up before 2nd toll booth?
• Time to “push” entire convoy through toll
  booth onto highway = 12*10 = 120 sec
• Time for last car to propagate from 1st to
  2nd toll both: 100km/(100km/hr) = 1 hr
• A: 62 minutes

Question 9

• Suppose cars now “propagate” faster - at 1000 km/hr
• And that toll booths now take one min to service a car
• Q: Will cars arrive at the 2nd booth before all cars have been serviced at the1st?

Answer 9

A: Yes! after 7 min, first car arrives at the 2nd booth; three cars are still at the 1st booth
– First car spends 1 minute at toll booth, and then 1/10 of an hour = 6 mins propagating through the link until arriving at 2nd booth
– At this point, car 8 is still at 1st booth (with 2 cars waiting behind)

Question 10

Nodal Delay

Answer 10

• Total of all previously-mentioned types of delay
• Measured per node (i.e., per each device in a network)

Question 11

End-to-End Delay and RTT

Answer 11

• End-to-End Delay is an often-used measurement that is the total of all nodal delays, from one host to another
  
  • May include numerous nodes
• Varies over time as the various component sources of delay increase and decrease
• Round Trip Time (RTT): End-to-End Delay measured in both directions
  
  • From one to host to another, and then back again
  • Doesn’t necessarily have to use the same path/route!

Question 12

Measuring Delay
Traceroute

Answer 12

• What does “real” Internet delay look like?
• The traceroute program measures delay from the source to each router on the path to a destination
• Records RTT (there and back)
• For each router, the source:
  – Sends three packets (probes) to a router that lies on the path to the destination
  – Routers return packets to the sender
  – Sender measures lag between transmission and reply

Question 13

Throughput Overview

Answer 13

• Throughput is the rate (bits/timeunit) at which bits are transferred from a sender to a receiver
  – Instantaneous throughput: rate at given point in time
  – Average throughput: rate over a period of time
  – Peak throughput: highest instantaneous throughput rate seen so far
• Throughput is often restricted by a single-point bottleneck
• Some protocols can “throttle” themselves, and reduce their own rate
  – This avoids stressing bottleneck, but at the cost of lower rates

Question 14

bottleneck link

Answer 14

link on the end-end path that constrains end-end throughput

Question 15

Throughput (Internet Scenario)

Answer 15

• Per-connection end-end throughput: min(Rc,Rs,R/10)
• In practice: Rc or Rs is often the bottleneck

Question 16

Goodput

Answer 16

• A term sometimes used when discussing throughput…
• Goodput measures throughput at the highest level and aims to be an ‘honest’ application-meaningful measurement
• It excludes protocol header and retransmission overheads

Question 17

Measuring throughput
iperf

Answer 17

• Produces standardised throughput measurements
• Normally runs as a “client-server” model
  
  • Client host requests sample data, server host serves it, generating a stream of sample data across the network
  • Because we know exactly how large the sample data is, and how long it takes to retrieve it, we can calculate throughput
• Very configurable
  
  • Can use different protocols, data sample sizes, etc.
  • Can also do bi-directional transfer, to test both directions
    * Check for irregularities

Question 18

Packet Loss Overview

Answer 18

• Packet Loss occurs when a router node drops or discards a packet
• This means that the packet does not reach its destination
• Packet Loss reduces the useable throughput (goodput) of a device, as the number of useful bits sent is fewer than would otherwise be the
  case
• For reliable protocols, packet loss causes retransmission to occur
  – The destination host sets a timer for the expected arrival time of each next-expected packet
  – If the timer expires before the packet arrives, a retransmission request is sent to the source host
  – This introduces additional delay, as the packet has to make the full journey again, regardless of where it was lost

Question 19

Queuing Delay and Packet Loss

Answer 19

• Packet Loss has a close relationship with Queuing Delay
• When packets arrive at a rate greater than the maximum supported throughput, the router’s queue starts to fill up

Question 20

Queuing Delay and Packet Loss 2

Answer 20

• When a queue becomes full, additional packets cannot be received
  
  • As a result, they are discarded and lost
• Lost packets may be retransmitted by the previous router, or by the source host, or not at all

Question 21

Other Sources of Loss

Answer 21

• Packets can also be lost in the physical medium, particularly with wireless link technologies
• Hardware and software in devices may also malfunction
  – This includes errors and corruption
  – packets may still be sent, but fail a checksum verification
• Devices can also be attacked: “Denial of Service” attack
  – A simple way would be to bombard a device with packets, and so fill up its buffers
  – This prevents other hosts from sending packets, whether completely or at a reduced rate

Question 22

Measuring Loss
ping

Answer 22

• Ping works in a similar fashion to traceroute but in an end-toend manner
• It creates messages to be sent out to specific hosts (rather than nodes along the path)
• Measures RTT (as with traceroute)
• Also includes a measurement of packet loss, by keeping track of how many messages were sent, and how many responses were received

Question 23

Queuing Disciplines

Answer 23

• When packets must be dropped because of a full queue, we still have control over which packet(s) to drop
• In the examples we have discussed, we have assumed a “tail drop” approach
• However, other techniques can be used:
  – Random Drop: Drop any packet within the queue
  – Quality-of-Service (QoS) Aware: Packets will be dropped given their priority; provide fairness and guarantee throughput for sensitive services, such as voice calls or live video