1
00:00:00,570 --> 00:00:03,950
Let's discuss ways to avoid congestion.

2
00:00:03,980 --> 00:00:11,210
The main mechanism that we'll discuss here is a W read or weighted random early detection w rate is

3
00:00:11,210 --> 00:00:13,800
a way to avoid congestion.

4
00:00:13,880 --> 00:00:21,110
But before we discuss w rate let's look at the problem cues on routers and switches are finite.

5
00:00:21,140 --> 00:00:27,860
In other words they can only hold or buffer a certain number of packets if there's a burst of traffic

6
00:00:28,100 --> 00:00:30,200
and the buffers are overrun.

7
00:00:30,200 --> 00:00:37,090
In other words there are more packets arriving than can be transmitted and buffered by a router with

8
00:00:37,100 --> 00:00:37,920
switch.

9
00:00:38,300 --> 00:00:43,470
It will started dropping packets when the queue folds up.

10
00:00:43,780 --> 00:00:46,660
All new packets arriving will be dropped.

11
00:00:46,660 --> 00:00:52,750
That's called tailed drop as an analogy think of a bucket.

12
00:00:53,120 --> 00:00:57,380
The bucket is filling up with water or packets.

13
00:00:57,530 --> 00:01:03,220
Once the bucket is full or in the case of a rod or a switch the queue is full.

14
00:01:03,230 --> 00:01:05,950
Any new packets that arrive are dropped.

15
00:01:07,170 --> 00:01:10,990
Because there's no space to buffer them or hold them in memory.

16
00:01:11,190 --> 00:01:20,400
Now when tail drop is used it results in wasted bandwidth especially when using TCE P here's an example

17
00:01:20,400 --> 00:01:27,590
of what happens when tailed drop is used and you've got lots of TCB flows the 100 percent is 100 percent

18
00:01:27,590 --> 00:01:35,480
utilization of an interface the y axis is utilization of an interface with 100 percent indicated by

19
00:01:35,480 --> 00:01:40,170
this line x axis is time.

20
00:01:40,230 --> 00:01:48,860
So what happens is if multiple flows start sending traffic and tell drop occurs at this point packets

21
00:01:48,860 --> 00:01:51,350
from all flows are dropped.

22
00:01:51,380 --> 00:01:58,450
Now what happens with TTP is that when packets are dropped the TTP senders slow down.

23
00:01:58,480 --> 00:02:06,040
In other words they back off they slow down and then they slowly increase the window size again to increase

24
00:02:06,040 --> 00:02:07,910
the speed of forwarding.

25
00:02:08,050 --> 00:02:13,540
So what happens is the utilization of interface goes down because the three centers back off.

26
00:02:13,630 --> 00:02:19,270
In other words they slow down they then increase they would decide to increase the number of packets

27
00:02:19,270 --> 00:02:22,810
that they are transmitting before getting an acknowledgement.

28
00:02:22,930 --> 00:02:29,950
But when the buffers are full again packets are dropped from all these flows including any new flows

29
00:02:30,430 --> 00:02:36,600
that start sending traffic so in this case we have four senders backing off or slowing down.

30
00:02:36,750 --> 00:02:41,400
Again they increase their window size and increase the number of packets that they can transmit before

31
00:02:41,400 --> 00:02:47,360
getting an acknowledgement and eventually it gets to the point where the buffers of a router or a switch

32
00:02:47,360 --> 00:02:48,710
are full again.

33
00:02:48,950 --> 00:02:53,260
Packets from all flows are dropped so they all slowed down.

34
00:02:53,510 --> 00:02:59,850
So the problem is is when the bucket is full packets from multiple flows are being received there's

35
00:02:59,870 --> 00:03:01,630
no space to buffer them.

36
00:03:01,990 --> 00:03:09,470
So packets are dropped from all the flows all centers back off at the same time and then increase speed

37
00:03:09,560 --> 00:03:11,240
at the same time.

38
00:03:11,500 --> 00:03:17,720
C end up having what's called a global synchronization where multiple centres are increasing their window

39
00:03:17,720 --> 00:03:19,050
size at the same time.

40
00:03:19,870 --> 00:03:25,610
They're slowing down at the same time increasing their speed at the same time slowing down at the same

41
00:03:25,610 --> 00:03:27,860
time and so forth and so on.

42
00:03:27,860 --> 00:03:33,260
Over time you never get full utilization of the interface.

43
00:03:33,300 --> 00:03:41,760
Now the idea with w red and congestion avoidance is you randomly drop packets from multiple flows before

44
00:03:42,270 --> 00:03:50,070
the queue folds up so in other words before the bucket is full or the cube is full you are ready dropping

45
00:03:50,100 --> 00:03:59,220
packets from flows but you do that randomly so you might drop a packet from flow one which means flow

46
00:03:59,220 --> 00:04:07,230
one will slow down but while center one is slowing down send it to is increasing its speed because its

47
00:04:07,230 --> 00:04:08,480
packets haven't been dropped.

48
00:04:09,860 --> 00:04:14,710
So while one host is slowing down another one is increasing its speed.

49
00:04:14,720 --> 00:04:20,600
This happens randomly so you've got some hosts increasing their speed and some slowing down at the same

50
00:04:20,600 --> 00:04:29,050
time instead of all of them slowing down and all of them speeding up at the same time so Cisco uses

51
00:04:29,050 --> 00:04:37,510
w red which introduces this randomness to allow better utilization of an interface interfaces bandwidth

52
00:04:38,080 --> 00:04:44,230
because some are slowing down and some are speeding up at the same time which in aggregate gives you

53
00:04:44,230 --> 00:04:50,760
a better utilization of the interface so the idea with W rate is you have a minimum threshold and a

54
00:04:50,760 --> 00:04:56,730
maximum threshold these valleys are below the size of the full queue.

55
00:04:57,940 --> 00:05:05,240
So the idea is when the average queue size or average queue depth is below the minimum threshold.

56
00:05:05,280 --> 00:05:12,730
Note packets have dropped when the average queue depth goes above the minimum threshold but is below

57
00:05:12,730 --> 00:05:14,510
the maximum threshold.

58
00:05:14,560 --> 00:05:22,580
We have random drops of packets but when it goes over the maximum threshold we have full drops of a

59
00:05:22,580 --> 00:05:25,510
traffic CLOs.

60
00:05:25,530 --> 00:05:33,060
The reason why we have W as in weighted random early detection is you can weight this based on different

61
00:05:33,060 --> 00:05:41,220
classes so you can have different maximum thresholds for different traffic classes.

62
00:05:41,260 --> 00:05:49,690
You may want to start dropping all FCP traffic before you drop HDP traffic so you can create a different

63
00:05:49,690 --> 00:05:56,830
minimum and maximum threshold so that certain traffic types are tailed dropped or fully dropped before

64
00:05:56,830 --> 00:06:06,280
other traffic times that can be based as an example on IP precedents or DCP so the idea with w red is

65
00:06:06,280 --> 00:06:11,680
we started dropping packets before the queue is full.

66
00:06:11,680 --> 00:06:19,360
We are avoiding congestion by pre selecting which packets get dropped and typically only want to drop

67
00:06:19,680 --> 00:06:26,680
DCP packets because TTP flows will read transmit so randomly dropping packets instead of tail dropping

68
00:06:26,680 --> 00:06:29,850
them avoids global synchronization.

69
00:06:30,160 --> 00:06:35,800
Different TTP flows are increasing their speed while others are slowing down so you get better utilization

70
00:06:35,800 --> 00:06:37,770
of a link.

71
00:06:37,970 --> 00:06:44,880
You also ensure that there is buffer space left for your voice packets so you will set your maximum

72
00:06:44,880 --> 00:06:53,290
threshold low enough so that FCP packets are fully dropped while ensuring that they still space left

73
00:06:53,290 --> 00:06:55,660
in the buffer for voice traffic.

74
00:06:57,150 --> 00:07:01,760
So in summary we have multiple quality of service mechanisms.

75
00:07:01,920 --> 00:07:04,220
We have classification and mocking.

76
00:07:04,470 --> 00:07:11,890
We have policing shaping and remarking we have congestion management or scheduling tools and we have

77
00:07:11,890 --> 00:07:18,670
a link specific tools such as link fragmentation and into leaving the issue a course is only an introduction

78
00:07:18,670 --> 00:07:20,260
to quality of service.

79
00:07:20,290 --> 00:07:25,930
Have a look at the quality of service s or indeed guide for more information about quality of service

80
00:07:25,960 --> 00:07:32,110
and good examples of how to apply quality of service on physical switches and routers.