Original subtitles

In its basic form, TCP may have a window size of 1 2

which means that for every segment transmitted by the sender 3

the receiver sends an acknowledgement for that individual segment. 4

This however slows down the throughput dramatically 5

because the sender cannot transmit anymore data 6

until it to receives acknowledgement of that single segment transmitted. 7

The throughput would be very low 8

depending on the round trip timer 9

between sending data and receiving the acknowledgement. 10

TCP however thus allow for greater window sizes 11

to allow for more segments 12

to be transmitted before receiving an acknowledgement. 12

The window is the number of data segments 13

the sender is allowed to send 14

without getting an acknowledgement from the receiver. 15

In this case we’ve set it to 1 16

that means when host A wants to send traffic to host B 17

it can send 1 segment because the window size is set to 1 18

host B, once it receives that segment, sends an acknowledgement. 19

In this example, let’s assume that host A, send the segment with the sequence number of 1 20

host B would acknowledge for segment 2. 21

Host A will then send segment 2 22

and host B, once successfully receiving that segment 23

will acknowledge or ACK for segment 3. 24

Host A will then send segment 3 to host B. 25

This process will then continue for the duration of the session 26

this is obviously very reliable however, the throughput is very low. 27

A would need to buffer outgoing segments 28

until it receive an acknowledgement for the segment transmitted. 29

now for argument sake 30

let’s assume that it takes 1 second for traffic 31

with the start with the small window size 32

and then exponentially increase the window size to gauge 33

the amount of data that the receiver can receive and what the network can handle. 34

You’ve probably notice this, when downloading a file from the internet 35

initially the download speed is slow 36

but then increases to a certain point over time. 37

This is because the window size initially are small 38

but then increases exponentially until a packet is dropped 39

or the receiving host cannot handle the amount of data it's receiving. 40

So you’ll notice initially, that the download speed is very slow 41

increases very quickly and then gets to a point 42

and then stays around that speed. 43

So once again, let’s assume 44

that the hosts in this example have a fixed window size of 3. 45

That means that A can send 3 segments before receiving an acknowledgement. 46

So in this case host A sends segment 1,2 and 3 47

host B acknowledges for segment 4 48

thus letting A know that it successfully received segment 1,2 and 3 49

host A then sends segment 4,5 and 6 because it has a fix window size of 3 50

and host B acknowledges for segment 7 51

thus letting A know, that it had received segments 4,5 and 6. 52

Remember with acknowledgements, acknowledge the next segment that you want to received 53

not the segment that you have already received. 80

So host B does not acknowledge for segment 6 54

but acknowledges segment 7. 55

As I've mentioned, a sliding window allows hosts to gauge 56

the amount of data the receiver can receive and what the network can handle. 57

So in this example let’s assume that the host are using a sliding window 58

the way that they determine what the network can handle 59

is when a packet is dropped by the network, the host will slow down. 60

This information is covered in a lot more detail 61

in courses that contain quality of service. 62

With this course just assume that when a packet gets drop 63

they reduced their window size dramatically 64

in brief, the window size is either the window granted to the sender by the receiver 65

or a calculated window called the congestion window or CWND 66

the congestion window or CWND is initially set to very low value 67

at connection establishment and then increases at an exponential rate. 68

For every lost segment, the congestion window is half 69

after lost segment has been successfully re-transmitted 70

the congestion window grows again until it reaches 71

a value half of the original congestion window 72

it then slows down its growth, using an algorithm called congestion avoidance. 73

It does exponentially grows up to half the original congestion window size 74

and then increases slowly at the linear rate. 75

In quality of service Weighted Random Early Detection or WRED 76

can be used improve efficiency of TCP transmissions across the link 77

as packets are randomly dropped from various flows or various sessions 78

going across an individual interface 79

rather than packets from multiple senders being drop at the same time. 80

this avoid an issue called global synchronization 81

where packets from multiple TCP sessions are drop at the same time 82

and therefore multiple host reduce their window size and slowdown at the same time 110

and then gradually increase their window size 83

and therefore their throughput at the same time. 84

So you have a many hosts slowing down and speeding up at the same time 85

with WRED some hosts will be slowing down 86

and other hosts will be increasing their throughput because or random drops. 87

Please refer to quality of service documentation for more detail about WRED 88

In this example let’s assume that we start with the initial window size of 3. 89

So A transmits 3 segments to B, however only segment 1 and 2 arrive at host B 90

segment 3 goes missing. 91

Host B may reduce its window size in this example to 2 and acknowledge segment 3. 92

Remember the acknowledgement is for the next packet that that hosts expects to receive 93

and because host B did not receive segment 3 94

it’s acknowledging successful receipt of segment 1 and 2. 95

Host A will re-transmit segment 3 and in this example send segment 4 96

but also request a window size of 3. 97

Both those segment are successfully received by host B 98

so host B will acknowledge for segment 5 99

but in this example, still wants a window size of 2. 100

Host A will only send 2 segments because the negotiated window size is 2 101

but A may still request that the window size be increase to 3. 102

With the sliding window, there is dynamic negotiation of a window size 103

and that window size may change dramatically during a session 104

depending on what the receiver can process and what the network can handle.