1
00:00:00,000 --> 00:00:03,840
[AUDIO LOGO]

2
00:00:03,840 --> 00:00:08,170

3
00:00:08,170 --> 00:00:10,870
So the first advanced feature
that we want to look at

4
00:00:10,870 --> 00:00:13,630
is BGP route dampening.

5
00:00:13,630 --> 00:00:16,090
Now the idea behind
route dampening

6
00:00:16,090 --> 00:00:18,910
is if we have a
neighbor or a customer,

7
00:00:18,910 --> 00:00:23,920
maybe another service provider
who's constantly advertising

8
00:00:23,920 --> 00:00:26,890
and retracting a
route, this is going

9
00:00:26,890 --> 00:00:31,600
to cause a lot of convergence
problems in our network,

10
00:00:31,600 --> 00:00:34,420
and possibly not even
our network, possibly

11
00:00:34,420 --> 00:00:35,990
the entire internet.

12
00:00:35,990 --> 00:00:38,620
So the idea behind
route dampening

13
00:00:38,620 --> 00:00:42,880
is if a route flaps
too many times,

14
00:00:42,880 --> 00:00:45,340
and we'll see how the
calculations are done

15
00:00:45,340 --> 00:00:49,750
and what we can adjust, but
based on that routes history

16
00:00:49,750 --> 00:00:53,230
of flapping, we can then
suppress that route,

17
00:00:53,230 --> 00:00:56,020
not advertise it
to anybody else,

18
00:00:56,020 --> 00:00:59,380
well, until it becomes
more well behaved.

19
00:00:59,380 --> 00:01:02,320
So that's the idea
behind route dampening.

20
00:01:02,320 --> 00:01:05,379
Let's jump in and take a
look at some of its details.

21
00:01:05,379 --> 00:01:08,140
And then, of course, we'll
jump into the command line

22
00:01:08,140 --> 00:01:11,580
and actually take a
look at how this works.

23
00:01:11,580 --> 00:01:14,450
So if we're going to talk
about BGP route dampening

24
00:01:14,450 --> 00:01:18,170
and suppressing flapping
routes, we probably

25
00:01:18,170 --> 00:01:21,480
need to start by defining
exactly what that route

26
00:01:21,480 --> 00:01:23,360
flap actually is.

27
00:01:23,360 --> 00:01:27,240
And quite simply, it's
a route, of course,

28
00:01:27,240 --> 00:01:33,020
that's being advertised
and withdrawn several times

29
00:01:33,020 --> 00:01:35,550
in a short period of time.

30
00:01:35,550 --> 00:01:38,150
So when we talk about
a short period of time,

31
00:01:38,150 --> 00:01:40,100
we're generally
talking about this

32
00:01:40,100 --> 00:01:45,500
in the time frame of minutes
to hours, not weeks or months.

33
00:01:45,500 --> 00:01:50,060
Now the reason I put a star
up here on several times,

34
00:01:50,060 --> 00:01:51,505
this is adjustable.

35
00:01:51,505 --> 00:01:52,880
And that's what
we're going to be

36
00:01:52,880 --> 00:01:56,550
looking at in this video, the
things that we can adjust.

37
00:01:56,550 --> 00:01:58,850
And they will
affect how quickly,

38
00:01:58,850 --> 00:02:02,630
how often these
routes get suppressed.

39
00:02:02,630 --> 00:02:04,100
Now remember part
of the reason we

40
00:02:04,100 --> 00:02:09,650
need this is that BGP was
designed specifically to handle

41
00:02:09,650 --> 00:02:12,480
a large number of prefixes.

42
00:02:12,480 --> 00:02:15,200
Now the sort of
side effects to that

43
00:02:15,200 --> 00:02:23,550
is this does entail sacrificing
convergent speed for stability.

44
00:02:23,550 --> 00:02:25,430
So that means that
our convergence isn't

45
00:02:25,430 --> 00:02:28,040
going to be that fast with BGP.

46
00:02:28,040 --> 00:02:30,320
Are there things we
can do to speed it up?

47
00:02:30,320 --> 00:02:32,630
Yes, but then we
have to start being

48
00:02:32,630 --> 00:02:34,370
concerned about resources.

49
00:02:34,370 --> 00:02:37,160
And we do have a video
about that coming up here

50
00:02:37,160 --> 00:02:38,820
in a little bit.

51
00:02:38,820 --> 00:02:43,470
But to help address this issue
of scalability and stability,

52
00:02:43,470 --> 00:02:46,100
there are several
default timers,

53
00:02:46,100 --> 00:02:48,740
default features
in BGP to support

54
00:02:48,740 --> 00:02:51,650
the goal of being stable.

55
00:02:51,650 --> 00:02:54,860
And route flap
dampening, which is

56
00:02:54,860 --> 00:02:57,830
sort of the official name
for what we're talking about

57
00:02:57,830 --> 00:03:00,150
is one of those features.

58
00:03:00,150 --> 00:03:03,950
So, what was trying to be
achieved with BGP route

59
00:03:03,950 --> 00:03:04,910
dampening?

60
00:03:04,910 --> 00:03:07,350
Well the first thing
was, of course,

61
00:03:07,350 --> 00:03:11,060
to reduce the router
processing load.

62
00:03:11,060 --> 00:03:14,910
If we have huge BGP
tables, this of course,

63
00:03:14,910 --> 00:03:18,620
is going to cause more
CPU load on the devices.

64
00:03:18,620 --> 00:03:22,130
And every time they get
advertised and withdrawn,

65
00:03:22,130 --> 00:03:24,750
we have to do it again.

66
00:03:24,750 --> 00:03:27,350
And while we're trying to
protect the local router,

67
00:03:27,350 --> 00:03:30,320
we should also be trying
to minimize these eBGP

68
00:03:30,320 --> 00:03:33,260
updates on the entire internet.

69
00:03:33,260 --> 00:03:35,845
As a service provider,
we need to make sure

70
00:03:35,845 --> 00:03:37,220
that we don't have
customers that

71
00:03:37,220 --> 00:03:40,550
are affecting other service
providers, and so on.

72
00:03:40,550 --> 00:03:42,530
And of course,
another big thing here

73
00:03:42,530 --> 00:03:47,930
is to not suppress routes
that flap occasionally.

74
00:03:47,930 --> 00:03:51,080
Again, we're not talking about a
customer that had a maintenance

75
00:03:51,080 --> 00:03:55,220
window and therefore their
route got withdrawn and then

76
00:03:55,220 --> 00:03:57,080
re advertised to us.

77
00:03:57,080 --> 00:03:59,570
Maybe they even did it twice.

78
00:03:59,570 --> 00:04:03,620
They reconfigured a router, and
that withdrew and re-advertised

79
00:04:03,620 --> 00:04:04,650
the prefix.

80
00:04:04,650 --> 00:04:07,100
And then they realized they
had to reload that router,

81
00:04:07,100 --> 00:04:10,190
and that withdrew and
re-advertised the prefix.

82
00:04:10,190 --> 00:04:12,390
That's not what
we're talking about.

83
00:04:12,390 --> 00:04:16,100
We're talking about routes that
have flapped numerous times.

84
00:04:16,100 --> 00:04:18,140
Generally speaking
with default numbers

85
00:04:18,140 --> 00:04:21,890
we'll see in just a minute,
it's three or more times just

86
00:04:21,890 --> 00:04:23,780
with default numbers.

87
00:04:23,780 --> 00:04:25,670
And of course, our
biggest design goal

88
00:04:25,670 --> 00:04:31,400
here was to suppress routes
that are likely to flap.

89
00:04:31,400 --> 00:04:32,690
How do we know?

90
00:04:32,690 --> 00:04:34,940
We're going to base
this on their history.

91
00:04:34,940 --> 00:04:37,910
Again, it's flapped
several times

92
00:04:37,910 --> 00:04:40,740
in a short period of time.

93
00:04:40,740 --> 00:04:43,260
That's what we're
looking to suppress.

94
00:04:43,260 --> 00:04:44,990
So just sort of a
design statement

95
00:04:44,990 --> 00:04:48,290
here that even with
some of those default

96
00:04:48,290 --> 00:04:53,100
timers that we have in
BGP, such as, for example,

97
00:04:53,100 --> 00:04:56,690
the advertisement
interval, which by default

98
00:04:56,690 --> 00:05:01,370
limits updates, and this is
only the eBGP peers by the way,

99
00:05:01,370 --> 00:05:06,570
but it limits them to only
be sent every 30 seconds.

100
00:05:06,570 --> 00:05:12,200
So there is, again, some latency
built into BGP on purpose

101
00:05:12,200 --> 00:05:15,710
to make sure that a route
flapping every 5 seconds

102
00:05:15,710 --> 00:05:19,550
doesn't just go crazy
in the BGP tables.

103
00:05:19,550 --> 00:05:23,960
But it's still a good idea
to implement route dampening

104
00:05:23,960 --> 00:05:25,790
for more severe conditions.

105
00:05:25,790 --> 00:05:27,830
This 30 second
delay is not going

106
00:05:27,830 --> 00:05:31,070
to help if our customer is
withdrawing and advertising

107
00:05:31,070 --> 00:05:33,930
their route every five minutes.

108
00:05:33,930 --> 00:05:36,560
So this is just sort of
solidifying that statement

109
00:05:36,560 --> 00:05:39,470
that, again, those
timers do help,

110
00:05:39,470 --> 00:05:43,910
but it is still best to
just remove or suppress

111
00:05:43,910 --> 00:05:47,990
that flapping route
until it becomes stable.

112
00:05:47,990 --> 00:05:51,530
That's the whole goal
behind route dampening.

113
00:05:51,530 --> 00:05:53,690
So as we jump into
discussing this,

114
00:05:53,690 --> 00:05:56,910
we need to start by
defining some terms

115
00:05:56,910 --> 00:05:59,150
so we can understand
route dampening.

116
00:05:59,150 --> 00:06:01,790
First is penalty.

117
00:06:01,790 --> 00:06:03,770
Penalty is the value
that's attached

118
00:06:03,770 --> 00:06:07,080
to the route when it flaps.

119
00:06:07,080 --> 00:06:08,780
This value is 1,000.

120
00:06:08,780 --> 00:06:12,920
And it's the only number in here
that we can't either directly

121
00:06:12,920 --> 00:06:14,990
or indirectly change.

122
00:06:14,990 --> 00:06:17,850
Next is the half life.

123
00:06:17,850 --> 00:06:23,390
This is defined as the decay
point at which the penalty is

124
00:06:23,390 --> 00:06:26,780
half of its previous value.

125
00:06:26,780 --> 00:06:29,480
This defaults to 15 minutes.

126
00:06:29,480 --> 00:06:31,970
This one we can change.

127
00:06:31,970 --> 00:06:34,400
Next is the suppress limit.

128
00:06:34,400 --> 00:06:37,100
This is simply the
value at which we will

129
00:06:37,100 --> 00:06:39,270
start suppressing the route.

130
00:06:39,270 --> 00:06:41,900
So when the route
will be suppressed,

131
00:06:41,900 --> 00:06:45,470
and this defaults to 2,000.

132
00:06:45,470 --> 00:06:47,960
Now I said that with
the default numbers,

133
00:06:47,960 --> 00:06:50,808
it has to flap at
least three times.

134
00:06:50,808 --> 00:06:53,100
And you might be looking at
this saying, wait a minute.

135
00:06:53,100 --> 00:06:57,560
It gets suppressed at 2,000,
and the penalty is 1,000.

136
00:06:57,560 --> 00:07:01,400
Doesn't that mean that it'll
be suppressed after two flaps?

137
00:07:01,400 --> 00:07:02,420
And the answer's no.

138
00:07:02,420 --> 00:07:06,680
You'll see why in just a moment
when we put all this together.

139
00:07:06,680 --> 00:07:09,500
Next is the reuse limit.

140
00:07:09,500 --> 00:07:11,090
This is where we're
going to start

141
00:07:11,090 --> 00:07:13,890
advertising that route again.

142
00:07:13,890 --> 00:07:18,710
And as you can see, it
has to go down to 750.

143
00:07:18,710 --> 00:07:21,000
Again, these are default values.

144
00:07:21,000 --> 00:07:24,470
These can all be adjusted,
except for the penalty.

145
00:07:24,470 --> 00:07:26,960
Next is the forget limit.

146
00:07:26,960 --> 00:07:30,170
This defaults to 375.

147
00:07:30,170 --> 00:07:35,610
But what it actually is, it's
one half of the reuse limit.

148
00:07:35,610 --> 00:07:37,610
So we don't set
this one directly.

149
00:07:37,610 --> 00:07:40,310
This one would get
set indirectly.

150
00:07:40,310 --> 00:07:43,340
And this is the point at
which the history of the route

151
00:07:43,340 --> 00:07:47,300
flapping is erased, and we're
no longer keeping track.

152
00:07:47,300 --> 00:07:51,350
So we basically forgive
all the past history.

153
00:07:51,350 --> 00:07:55,150
And it wipes the slate
clean, and we start it over.

154
00:07:55,150 --> 00:07:57,880
Then we have the
max suppress time.

155
00:07:57,880 --> 00:08:01,810
If a customers route were
to flap over and over

156
00:08:01,810 --> 00:08:04,390
and over and over
again, and this penalty

157
00:08:04,390 --> 00:08:08,530
keeps going up 1,000,
plus 1,000, plus 1,000,

158
00:08:08,530 --> 00:08:13,330
plus 1,000, if this were just
to keep going, at some point,

159
00:08:13,330 --> 00:08:18,160
if we think it through, it
would take us a long time

160
00:08:18,160 --> 00:08:20,800
to get back down
to this 750 where

161
00:08:20,800 --> 00:08:23,330
we're going to actually
advertise the route again.

162
00:08:23,330 --> 00:08:29,200
So this is the maximum time
regardless of the penalty.

163
00:08:29,200 --> 00:08:31,750
No matter how high
that penalty got,

164
00:08:31,750 --> 00:08:35,320
when we get to the
maximum time, the router

165
00:08:35,320 --> 00:08:38,299
will simply re-advertise
the route again.

166
00:08:38,299 --> 00:08:41,200
So now that we have an idea
of what these numbers are

167
00:08:41,200 --> 00:08:46,880
and using the defaults, let's
take a look at this on a chart.

168
00:08:46,880 --> 00:08:50,590
So here on our chart, we
have penalty mapped up

169
00:08:50,590 --> 00:08:51,770
the left side.

170
00:08:51,770 --> 00:08:54,800
We have time being
shown across the bottom.

171
00:08:54,800 --> 00:08:58,120
And we have our various
limits shown over here

172
00:08:58,120 --> 00:09:03,470
on the right with their default
values over here on the left.

173
00:09:03,470 --> 00:09:05,320
So let's put this
together and come up

174
00:09:05,320 --> 00:09:09,160
with an example of what would
happen in the case of a route

175
00:09:09,160 --> 00:09:10,340
flapping.

176
00:09:10,340 --> 00:09:13,900
So we'll just say
here at time 0,

177
00:09:13,900 --> 00:09:16,510
we have a route actually flap.

178
00:09:16,510 --> 00:09:19,720
And that means it's going to
go to a penalty of 1,000, which

179
00:09:19,720 --> 00:09:21,470
is right about there.

180
00:09:21,470 --> 00:09:24,790
Now here's the thing,
what's going to happen

181
00:09:24,790 --> 00:09:30,100
is the router is now going
to start the half-life timer.

182
00:09:30,100 --> 00:09:36,490
And it's going to look to
times 0 plus 15 minutes.

183
00:09:36,490 --> 00:09:39,730
And it's going to start
drawing a line to where

184
00:09:39,730 --> 00:09:45,660
we would have to be at one half
the current value at that time

185
00:09:45,660 --> 00:09:46,970
frame.

186
00:09:46,970 --> 00:09:50,450
So it's going to start
drawing an imaginary line

187
00:09:50,450 --> 00:09:56,130
and start decaying our value
towards that half life.

188
00:09:56,130 --> 00:10:02,240
So let's say now the
route flaps again.

189
00:10:02,240 --> 00:10:07,810
So we'll say right here at
timestamp 1, the route flaps

190
00:10:07,810 --> 00:10:09,050
again.

191
00:10:09,050 --> 00:10:13,450
So now it gets
another 1,000 penalty.

192
00:10:13,450 --> 00:10:17,770
And we see right away why
it's not going to get dampened

193
00:10:17,770 --> 00:10:20,200
if it only flaps twice.

194
00:10:20,200 --> 00:10:23,260
Because even if this
decay is only a 1,

195
00:10:23,260 --> 00:10:27,460
even if it only drops it
to 999, the second flap

196
00:10:27,460 --> 00:10:31,053
is still only going
to bring it to 1,999.

197
00:10:31,053 --> 00:10:32,470
And by the way,
the routes are not

198
00:10:32,470 --> 00:10:34,210
going to flap that
quickly anyway

199
00:10:34,210 --> 00:10:37,150
because again we have
30-second delay timers

200
00:10:37,150 --> 00:10:38,660
and things like this.

201
00:10:38,660 --> 00:10:42,820
So it's going to have decayed
the penalty at least a little

202
00:10:42,820 --> 00:10:45,920
before it could even
flap a second time.

203
00:10:45,920 --> 00:10:48,580
But now that it's
flapped a second time,

204
00:10:48,580 --> 00:10:51,100
the first half life timer
is going to be removed,

205
00:10:51,100 --> 00:10:53,060
and then we're going
to make a new one.

206
00:10:53,060 --> 00:10:55,630
So we're going to go out,
again, I'm estimating here.

207
00:10:55,630 --> 00:11:00,940
We'll say that this is going
to be now time 1 plus 15.

208
00:11:00,940 --> 00:11:06,170
And now the half time is
probably right around there.

209
00:11:06,170 --> 00:11:08,120
So now what we're going
to do, once again,

210
00:11:08,120 --> 00:11:12,310
is start decaying
towards that half mark.

211
00:11:12,310 --> 00:11:16,660
So you can see because of the
way, this is a decay mechanism,

212
00:11:16,660 --> 00:11:19,930
the higher the penalty,
the faster it's actually

213
00:11:19,930 --> 00:11:22,510
going to decay the penalty.

214
00:11:22,510 --> 00:11:28,000
So again, though, if it
flaps yet again at time 2,

215
00:11:28,000 --> 00:11:31,480
now we're going to add
another 1,000 to the penalty.

216
00:11:31,480 --> 00:11:35,320
And at this point, we are
now going to go suppressed.

217
00:11:35,320 --> 00:11:37,670
So now, what happens?

218
00:11:37,670 --> 00:11:40,150
We get rid of our
half-life timer.

219
00:11:40,150 --> 00:11:43,000
Once again, we'll come
out here give or take.

220
00:11:43,000 --> 00:11:47,050
This is going to
be time 2 plus 15.

221
00:11:47,050 --> 00:11:50,662
And half is probably going
to be right around there.

222
00:11:50,662 --> 00:11:52,120
And what we're
going to do is we're

223
00:11:52,120 --> 00:11:56,890
going to start decaying this
penalty towards the half-life

224
00:11:56,890 --> 00:11:57,850
timer.

225
00:11:57,850 --> 00:12:00,890
This time, we'll say
it doesn't flap again.

226
00:12:00,890 --> 00:12:03,190
So now it's stable.

227
00:12:03,190 --> 00:12:06,710
We're back below the
suppressed limit.

228
00:12:06,710 --> 00:12:10,070
However, we're not down
to the reuse limit.

229
00:12:10,070 --> 00:12:12,070
So then, what happens?

230
00:12:12,070 --> 00:12:14,380
Since we've reached
this half-life timer,

231
00:12:14,380 --> 00:12:16,370
it's going to go away.

232
00:12:16,370 --> 00:12:18,500
We're going to put
another one out here,

233
00:12:18,500 --> 00:12:21,700
which is going to
be T2, essentially,

234
00:12:21,700 --> 00:12:23,920
it's going to be T2 plus 30.

235
00:12:23,920 --> 00:12:27,130
And now it's going to
be half of what remains.

236
00:12:27,130 --> 00:12:29,870
So we'll say somewhere
right around there.

237
00:12:29,870 --> 00:12:34,670
And now what happens is we
start decaying towards that.

238
00:12:34,670 --> 00:12:37,420
But as soon as we
cross that reuse limit,

239
00:12:37,420 --> 00:12:40,340
we start advertising it again.

240
00:12:40,340 --> 00:12:43,840
So at that point right
there, this route

241
00:12:43,840 --> 00:12:46,760
will then start getting
advertised again.

242
00:12:46,760 --> 00:12:51,680
However, we are still keeping
its history, which means--

243
00:12:51,680 --> 00:12:53,680
and we're not going to
follow through with this,

244
00:12:53,680 --> 00:12:57,850
but it means that if
this were to flap again,

245
00:12:57,850 --> 00:13:00,910
it would add the 1,000
penalty from here.

246
00:13:00,910 --> 00:13:03,730
It wouldn't start
from the bottom

247
00:13:03,730 --> 00:13:06,040
like we did the first time.

248
00:13:06,040 --> 00:13:08,530
It would pick up right
where it left off.

249
00:13:08,530 --> 00:13:10,780
It may not go right
back to suppressed,

250
00:13:10,780 --> 00:13:14,020
but it still has its
existing history.

251
00:13:14,020 --> 00:13:17,470
But this half-life
timer continues.

252
00:13:17,470 --> 00:13:25,030
Somewhere out here is going
to be a timer time 2 plus 45

253
00:13:25,030 --> 00:13:28,150
because it's been
three half-life timers.

254
00:13:28,150 --> 00:13:30,490
So now what happens is
we choose half of that.

255
00:13:30,490 --> 00:13:32,000
And just for the
sake of simplicity,

256
00:13:32,000 --> 00:13:34,300
I'm going to actually draw
it just below that line.

257
00:13:34,300 --> 00:13:36,640
That may or may not be
completely accurate.

258
00:13:36,640 --> 00:13:39,130
This is all just drawn freehand.

259
00:13:39,130 --> 00:13:43,060
But now we start decaying
towards that mark

260
00:13:43,060 --> 00:13:45,830
until we get to right there.

261
00:13:45,830 --> 00:13:48,590
And as soon as we
get right here,

262
00:13:48,590 --> 00:13:52,720
now everything's forgiven,
and this whole process

263
00:13:52,720 --> 00:13:53,830
would start over.

264
00:13:53,830 --> 00:13:57,490
So the slates wiped clean,
there's no more history,

265
00:13:57,490 --> 00:14:01,090
and the route would
start back here again

266
00:14:01,090 --> 00:14:03,530
if it were to flap later.

267
00:14:03,530 --> 00:14:05,860
So hopefully this
gives us a good idea

268
00:14:05,860 --> 00:14:09,950
of how BGP route
dampening actually works.

269
00:14:09,950 --> 00:14:13,690
So in our next video, let's
jump into the command line

270
00:14:13,690 --> 00:14:18,460
and take a look at how
this actually gets built.

271
00:14:18,460 --> 00:14:22,290
So in this video, we got an
introduction to BGP route

272
00:14:22,290 --> 00:14:25,950
dampening, why we would
use it, what it's for,

273
00:14:25,950 --> 00:14:28,560
and how the numbers
work together

274
00:14:28,560 --> 00:14:32,790
to create this chart that
explains when the routes get

275
00:14:32,790 --> 00:14:36,790
suppressed, when they get
unsuppressed, and then finally,

276
00:14:36,790 --> 00:14:39,960
when the history is removed
so that they start over

277
00:14:39,960 --> 00:14:43,950
with a zero penalty instead
of picking up where they were.

278
00:14:43,950 --> 00:14:46,230
Again, in the next
video, we're going

279
00:14:46,230 --> 00:14:50,160
to jump into the command line
and actually see this working.

280
00:14:50,160 --> 00:14:52,350
I hope this has been
informative for you.

281
00:14:52,350 --> 00:14:55,490
And I'd like to thank
you for viewing.

282
00:14:55,490 --> 00:14:59,000