1
00:00:00,060 --> 00:00:06,810
‫Today's topic is right, amplification, database, right, amplification, you know, guys, this is

2
00:00:06,810 --> 00:00:15,390
‫one of the reasons that Ueber have moved from PostgreSQL to my school and we obviously talked about

3
00:00:15,630 --> 00:00:23,520
‫that move that is not entirely approved by the community because nobody missed a lot of things in that

4
00:00:23,520 --> 00:00:27,490
‫article, but it was a valid reason nevertheless.

5
00:00:27,840 --> 00:00:28,050
‫So.

6
00:00:28,050 --> 00:00:28,260
‫Right.

7
00:00:28,260 --> 00:00:36,570
‫Amplification, if you don't know guys, it's like a phenomena where your update stadtmueller your school

8
00:00:36,570 --> 00:00:37,290
‫or whatever.

9
00:00:37,290 --> 00:00:37,710
‫Right.

10
00:00:37,710 --> 00:00:39,330
‫You are performing logically.

11
00:00:39,840 --> 00:00:42,870
‫The actual physical representation of that.

12
00:00:42,870 --> 00:00:50,640
‫Right, is double or triple or maybe a magnitude more order of magnitude more than the actual logical.

13
00:00:50,640 --> 00:00:50,950
‫Right.

14
00:00:51,210 --> 00:00:55,230
‫So you would think like, hey, I just did a single field in a single role.

15
00:00:55,440 --> 00:01:03,870
‫But you look at the disk usage at that point and you see it doing thousands and thousands and thousands

16
00:01:03,870 --> 00:01:04,920
‫of kilobytes.

17
00:01:04,920 --> 00:01:11,820
‫Like what what I did, I, I merely updated a 32 bit integer a what the heck happened?

18
00:01:12,090 --> 00:01:13,350
‫The right amplification.

19
00:01:13,350 --> 00:01:15,480
‫You can look at it from three tiers.

20
00:01:15,690 --> 00:01:19,620
‫The first thing was I kind of I saw from building apps.

21
00:01:19,620 --> 00:01:19,940
‫Right.

22
00:01:20,160 --> 00:01:21,720
‫And I called it the application.

23
00:01:21,720 --> 00:01:22,050
‫Right.

24
00:01:22,050 --> 00:01:22,950
‫Amplification.

25
00:01:23,160 --> 00:01:25,320
‫The second type is the data base.

26
00:01:25,320 --> 00:01:25,650
‫Right.

27
00:01:25,650 --> 00:01:28,920
‫Amplification and the third time is the disk right.

28
00:01:28,920 --> 00:01:30,750
‫Amplification, which is the SSD.

29
00:01:30,750 --> 00:01:31,860
‫Right amplification.

30
00:01:32,040 --> 00:01:34,650
‫I'm going to talk about the three of them as this.

31
00:01:34,710 --> 00:01:37,740
‫These are infamous for right amplification.

32
00:01:37,740 --> 00:01:39,060
‫So I'm going to talk about that.

33
00:01:39,060 --> 00:01:43,260
‫And then database, which is something that is hidden from you, depends really.

34
00:01:43,260 --> 00:01:48,650
‫Database platform and then the application itself, which is kind of your fault as a backing engineer.

35
00:01:48,990 --> 00:01:50,640
‫So let's jump onto each one of them.

36
00:01:50,830 --> 00:01:51,020
‫All right.

37
00:01:51,300 --> 00:01:52,890
‫So let's start with the application.

38
00:01:53,040 --> 00:01:53,520
‫Right.

39
00:01:53,520 --> 00:02:02,640
‫Amplification and this is maybe your end user is expecting something, but it turns out that the actual

40
00:02:02,640 --> 00:02:10,970
‫work behind that API single call or update is magnificently larger.

41
00:02:11,400 --> 00:02:15,360
‫For example, you're taking your to do item.

42
00:02:15,360 --> 00:02:22,530
‫Hey, I just done this desk, but so to the to the developer, the front end developer in this case

43
00:02:22,920 --> 00:02:31,390
‫calling that API to kind of just hey, I just deleted I just I just marked this as done right so far.

44
00:02:31,440 --> 00:02:31,890
‫The developer.

45
00:02:31,890 --> 00:02:32,340
‫My thing.

46
00:02:32,340 --> 00:02:33,510
‫Oh this is just an update.

47
00:02:33,510 --> 00:02:39,900
‫Statement maybe our own a Synagro, but a back end developer could have written this application in

48
00:02:39,900 --> 00:02:48,540
‫a way so that, hey, it goes to the to do app mark that's to do as done and then goes to another table,

49
00:02:48,690 --> 00:02:53,430
‫maybe inserts that raw into another archive table.

50
00:02:53,430 --> 00:02:54,870
‫Hey there, Hosain.

51
00:02:54,870 --> 00:02:57,660
‫Have done has washed the dishes, for example.

52
00:02:57,980 --> 00:02:58,920
‫It goes into another.

53
00:02:58,920 --> 00:02:59,580
‫So that's two.

54
00:02:59,580 --> 00:03:00,060
‫Right.

55
00:03:00,270 --> 00:03:07,290
‫And then maybe you have another table maintaining the count of the task that you have done.

56
00:03:07,290 --> 00:03:12,200
‫So you your increment or your update, another table with the actual work.

57
00:03:12,210 --> 00:03:16,080
‫So one single API translate to many, many rides.

58
00:03:16,080 --> 00:03:19,160
‫And obviously there's this goes downstream and affects us.

59
00:03:19,350 --> 00:03:22,110
‫So this is the simplest of thing you think is the single.

60
00:03:22,110 --> 00:03:29,010
‫Right, but the actual API implementation, it does a lot of things and relational databases are designed

61
00:03:29,010 --> 00:03:34,530
‫this way, especially if you have normalized table where you have many, many tables that you have to

62
00:03:34,860 --> 00:03:37,440
‫kind of maintain and keep it consistent.

63
00:03:38,140 --> 00:03:38,490
‫Right.

64
00:03:39,180 --> 00:03:42,630
‫That doesn't mean obviously N Equal is infamous of that.

65
00:03:42,630 --> 00:03:45,270
‫Obviously, it has its own limitation as well.

66
00:03:45,630 --> 00:03:48,480
‫But now this is the first time the application.

67
00:03:48,780 --> 00:03:49,440
‫Right.

68
00:03:49,440 --> 00:03:55,800
‫Amplification your app claims to do one thing, but actually behind the scene is doing way, way, way,

69
00:03:55,800 --> 00:03:56,760
‫way, way, way more.

70
00:03:57,190 --> 00:04:04,770
‫Another famous thing is delete some people, implement, delete as as a as an actual update and an insert.

71
00:04:04,770 --> 00:04:08,820
‫Some people implement delete as as a logical delete.

72
00:04:08,820 --> 00:04:09,120
‫Right.

73
00:04:09,120 --> 00:04:11,250
‫Where do they do more and more work at the back end.

74
00:04:11,370 --> 00:04:12,920
‫So a really different.

75
00:04:12,940 --> 00:04:13,950
‫So that's an application.

76
00:04:13,950 --> 00:04:14,130
‫Right.

77
00:04:14,790 --> 00:04:19,080
‫Let's talk about database write ID and this is whenever you hear right.

78
00:04:19,080 --> 00:04:25,350
‫Amplification, it could be either the the database ID or the other one which is this is the ID.

79
00:04:26,100 --> 00:04:32,070
‫So the database ID is is a property that is hidden from you as a database platform.

80
00:04:32,430 --> 00:04:38,850
‫I'm going to give the example of positives because of the most infamous and it is the exact same thing

81
00:04:38,880 --> 00:04:41,220
‫as my sequel.

82
00:04:41,760 --> 00:04:49,650
‫But with that, my isaam storage engine, they both work almost identically in this case.

83
00:04:49,650 --> 00:04:50,850
‫Oh, not not quite.

84
00:04:50,850 --> 00:04:55,980
‫So let's say you have a table with six columns, write five columns, have an index indexes, each has

85
00:04:55,980 --> 00:04:56,700
‫an index on them.

86
00:04:56,970 --> 00:04:59,940
‫And the last one, which is the sixth column, doesn't happen in.

87
00:05:00,380 --> 00:05:09,380
‫So if you execute an updated statement, you pulled a singer on the stable and then you updated one

88
00:05:09,380 --> 00:05:13,890
‫of the columns, I say column one, which happened to have an index on them in PostgreSQL.

89
00:05:14,240 --> 00:05:15,230
‫This is what will happen.

90
00:05:15,710 --> 00:05:18,600
‫Postgres will create a brand new row.

91
00:05:18,620 --> 00:05:19,490
‫That's how it works.

92
00:05:19,490 --> 00:05:25,390
‫Always any update, any modification on the Tupelo's create a new Topal idea.

93
00:05:25,400 --> 00:05:28,690
‫That is a terrible idea that uniquely identifies the road.

94
00:05:29,090 --> 00:05:32,890
‫Yeah, it will create a new one with the new values that you have.

95
00:05:33,620 --> 00:05:34,780
‫Copy, exact copy.

96
00:05:35,030 --> 00:05:38,520
‫We have to update the column responding index because we choose the value.

97
00:05:38,540 --> 00:05:40,540
‫So now we update the index.

98
00:05:40,940 --> 00:05:43,250
‫That's something we have to do anyway.

99
00:05:43,490 --> 00:05:44,470
‫So that's another.

100
00:05:44,470 --> 00:05:44,670
‫Right.

101
00:05:44,670 --> 00:05:52,700
‫So we inserted a new role and now we want to go to the index and change that tuple idea from pointing

102
00:05:52,700 --> 00:05:55,250
‫to the older or it points to the new role.

103
00:05:55,250 --> 00:05:57,110
‫And we also change the value.

104
00:05:57,110 --> 00:06:01,840
‫We introduce the new value that you inserted and that tuple in that right.

105
00:06:01,880 --> 00:06:04,780
‫So now it's almost two changes.

106
00:06:04,850 --> 00:06:07,340
‫And you might say, what about the rest of the index?

107
00:06:07,460 --> 00:06:09,510
‫These five indexes still point to the old row.

108
00:06:10,220 --> 00:06:15,530
‫Polska is used to update those indexes, to point to the new tuple.

109
00:06:16,280 --> 00:06:19,430
‫And that's what you didn't like back in, what, 91 or 92?

110
00:06:19,430 --> 00:06:20,110
‫I don't remember.

111
00:06:20,690 --> 00:06:28,940
‫And so so the right volume volume that is happening as a result was enormous, especially when you start

112
00:06:29,240 --> 00:06:30,940
‫you did a single update.

113
00:06:30,960 --> 00:06:33,830
‫Now you're touching all these index and you have a lot of indexes.

114
00:06:33,830 --> 00:06:38,570
‫All of sudden these indexes have to be touch despite them not being changed.

115
00:06:38,570 --> 00:06:40,550
‫If you think about it, they will not change.

116
00:06:40,850 --> 00:06:43,730
‫But Postgres actually solved this problem.

117
00:06:43,740 --> 00:06:48,680
‫I don't remember when maybe it was nine four or nine three doing something called the heap only to build

118
00:06:48,680 --> 00:06:49,790
‫a hot Topal.

119
00:06:50,900 --> 00:06:58,730
‫That essentially it's a trick, so, OK, it says we didn't really change the value if we didn't really

120
00:06:58,730 --> 00:07:04,400
‫change a value, we don't really go to the index to update it because that's that's just wasteful.

121
00:07:05,240 --> 00:07:07,460
‫So they they definitely fixed that.

122
00:07:07,490 --> 00:07:15,500
‫The only Topal will create the road, the new road to topple it and then points to the old road to the

123
00:07:15,500 --> 00:07:21,860
‫new one is that, hey, by the way, if anyone happened to land on the old road, tell them that there

124
00:07:21,860 --> 00:07:23,210
‫is a new version of this.

125
00:07:23,210 --> 00:07:28,970
‫There is a new heap, only Topal that points the only limitation here.

126
00:07:28,980 --> 00:07:30,830
‫It has to be in the same page.

127
00:07:30,830 --> 00:07:31,100
‫Right.

128
00:07:31,220 --> 00:07:34,760
‫If the page is full, then you cannot really do that.

129
00:07:34,760 --> 00:07:35,000
‫Right.

130
00:07:35,010 --> 00:07:36,760
‫That's that's why it's called hipbone and that's why.

131
00:07:36,780 --> 00:07:38,420
‫Well, full factor come in.

132
00:07:38,570 --> 00:07:40,530
‫You have to choose the correct fill of factors.

133
00:07:40,550 --> 00:07:42,500
‫It's a really complex things to understand.

134
00:07:42,510 --> 00:07:46,570
‫You have to be really careful when you're building this stuff anyway.

135
00:07:46,790 --> 00:07:49,700
‫So it used to do that, but no, not anymore.

136
00:07:49,760 --> 00:07:51,530
‫But I wanted to give this example to right.

137
00:07:51,530 --> 00:07:57,940
‫Amplification is a faint, especially with Postgres, but it could be with any really any other database.

138
00:07:58,100 --> 00:08:00,680
‫Let's say you did another one single update.

139
00:08:01,220 --> 00:08:03,140
‫You updated one roll with many values.

140
00:08:03,140 --> 00:08:05,000
‫You changed almost all columns.

141
00:08:05,120 --> 00:08:05,450
‫Right.

142
00:08:06,080 --> 00:08:12,560
‫There is no way in my school, on Postgres, on Oracle, we have to do we wrote one single value, one

143
00:08:12,560 --> 00:08:16,210
‫single rule, but as a result, we have to touch all the indexes.

144
00:08:16,400 --> 00:08:17,100
‫That's another.

145
00:08:17,100 --> 00:08:17,410
‫Right.

146
00:08:17,420 --> 00:08:24,530
‫Amplification, OK, because if you look at the update segment, it's logically, it looks to me like

147
00:08:24,530 --> 00:08:25,730
‫updated a single row.

148
00:08:26,090 --> 00:08:29,990
‫But to the database, you actually did more and more work.

149
00:08:29,990 --> 00:08:33,620
‫You did a lot a lot of physical work.

150
00:08:34,160 --> 00:08:37,280
‫And when I say physical here, I mean database physical.

151
00:08:37,280 --> 00:08:37,580
‫Right.

152
00:08:37,580 --> 00:08:43,340
‫Because it gets really confusing when I say logical and physical, where ordinarily when I say logical

153
00:08:43,340 --> 00:08:44,120
‫database.

154
00:08:44,630 --> 00:08:45,290
‫Right.

155
00:08:45,290 --> 00:08:51,620
‫Is the school and physical database right on essentially the wall changes the right ahead log.

156
00:08:51,660 --> 00:08:51,920
‫Right.

157
00:08:52,070 --> 00:08:54,690
‫All these actual changes that you have the pages.

158
00:08:54,710 --> 00:08:54,970
‫Right.

159
00:08:55,580 --> 00:09:01,880
‫But when it comes to this, this D, which is the next topic, that's a different beast, physical and

160
00:09:01,880 --> 00:09:04,550
‫logical becomes different things right now.

161
00:09:04,790 --> 00:09:07,430
‫This is the database ID, very powerful.

162
00:09:07,430 --> 00:09:08,660
‫You have to really understand it.

163
00:09:08,660 --> 00:09:08,880
‫Right.

164
00:09:09,080 --> 00:09:14,540
‫So that's the second type of medication that goes to the final one, which is a very infamous one without

165
00:09:14,540 --> 00:09:15,200
‫ID.

166
00:09:15,410 --> 00:09:19,190
‫Even if you go to the wiki page, when you write random application, that's the only one that shows

167
00:09:19,190 --> 00:09:22,160
‫up anyway, which is the zero ID.

168
00:09:23,180 --> 00:09:30,290
‫So to talk about that, we really need to understand how is this really works.

169
00:09:30,590 --> 00:09:38,180
‫So this is the guys have this idea of someone invented something called the charge strap is a technique

170
00:09:38,180 --> 00:09:41,630
‫where you can store multiple level of electrons.

171
00:09:42,020 --> 00:09:50,690
‫You can from zero electrons up to certain levels and the quantity of the electrons determines certain

172
00:09:50,690 --> 00:09:51,110
‫values.

173
00:09:51,110 --> 00:09:54,260
‫And this is the trick we used to correspond to this.

174
00:09:54,410 --> 00:10:00,770
‫And these trap are put into cells like so Acel can have a lot of charge traps.

175
00:10:01,130 --> 00:10:09,980
‫And as a result, one single cell can have up to the recent newer stuff up to 16 different charge levels.

176
00:10:10,370 --> 00:10:14,720
‫So if you do the math right with the computer science, 16 is how many bits?

177
00:10:14,720 --> 00:10:15,410
‫It's four bits.

178
00:10:15,410 --> 00:10:15,710
‫Right.

179
00:10:16,310 --> 00:10:21,530
‫So that means each cell can store four bits and once you charge them, you store them.

180
00:10:22,280 --> 00:10:25,160
‫You can turn off electricity works.

181
00:10:25,160 --> 00:10:26,150
‫It's persisted.

182
00:10:26,150 --> 00:10:27,800
‫It stays for ages.

183
00:10:27,950 --> 00:10:31,510
‫It does not go it does not get released.

184
00:10:31,520 --> 00:10:31,850
‫Right.

185
00:10:32,270 --> 00:10:34,880
‫And you can essentially read those values back.

186
00:10:35,060 --> 00:10:39,170
‫And then based on the quantity of number of electrons, they'll get this.

187
00:10:39,170 --> 00:10:40,580
‫Oh, this is zero zero zero zero.

188
00:10:40,700 --> 00:10:42,140
‫Oh, this is zero zero zero one.

189
00:10:42,140 --> 00:10:43,580
‫Oh, this is zero zero one zero.

190
00:10:43,580 --> 00:10:44,100
‫This is oh.

191
00:10:44,100 --> 00:10:48,700
‫Is this the one one one one based on this trick, they build the entire SSD.

192
00:10:48,800 --> 00:10:53,840
‫So now they took one cell, they organized into a row and then the rose becomes the, the, the, the

193
00:10:53,960 --> 00:10:54,590
‫pages.

194
00:10:54,590 --> 00:10:54,770
‫Right.

195
00:10:54,770 --> 00:10:57,380
‫So there is a page has many cells and many rows.

196
00:10:57,650 --> 00:11:06,080
‫I so think of a page now this have a lot of these charge drives so a page can have a lot of bits stored

197
00:11:06,080 --> 00:11:07,310
‫in it and you can do that.

198
00:11:07,490 --> 00:11:10,010
‫And then pages are organized into blocks.

199
00:11:10,010 --> 00:11:10,280
‫Right.

200
00:11:10,430 --> 00:11:11,390
‫And here's the trick here.

201
00:11:11,810 --> 00:11:12,130
‫Right.

202
00:11:12,170 --> 00:11:16,130
‫So now you have you have control of a block that has many pages.

203
00:11:16,130 --> 00:11:18,500
‫And this is what we're interested in here at this level.

204
00:11:18,500 --> 00:11:18,770
‫Right.

205
00:11:19,040 --> 00:11:23,750
‫And if you love to read more about it, I'm going to reference a video that absolutely offers a very

206
00:11:23,750 --> 00:11:25,790
‫technical, low level thing.

207
00:11:25,850 --> 00:11:27,200
‫If you're going to learn more about it.

208
00:11:27,200 --> 00:11:27,700
‫This is deep.

209
00:11:28,250 --> 00:11:29,750
‫It's a very good animations.

210
00:11:29,750 --> 00:11:30,530
‫So I loved it.

211
00:11:31,460 --> 00:11:34,430
‫So so now this is what we need to understand.

212
00:11:34,430 --> 00:11:36,290
‫Block and pages.

213
00:11:36,800 --> 00:11:38,420
‫However, nothing is free.

214
00:11:38,420 --> 00:11:40,370
‫Obviously, here is the condition.

215
00:11:40,720 --> 00:11:41,420
‫Is this D?

216
00:11:41,630 --> 00:11:42,050
‫Yeah.

217
00:11:42,470 --> 00:11:46,190
‫You can write stuff for us, you can write in pages.

218
00:11:47,060 --> 00:11:47,360
‫Right.

219
00:11:47,360 --> 00:11:49,730
‫As long as the page is.

220
00:11:50,800 --> 00:11:58,150
‫It doesn't have anything right, does it have any levels kept as one of the pages you can write anything

221
00:11:58,150 --> 00:11:59,240
‫you want on it, right?

222
00:11:59,320 --> 00:12:00,940
‫And when you write it, we're going to store it.

223
00:12:01,360 --> 00:12:08,800
‫But if there is a page that is already has some content in it and you want to update that content to

224
00:12:08,800 --> 00:12:12,160
‫something else, let's say you have a bunch of rows and have values.

225
00:12:12,250 --> 00:12:17,160
‫You want to update something, you have to erase this page and write it back.

226
00:12:17,350 --> 00:12:19,470
‫You cannot just overwrite it, right.

227
00:12:20,260 --> 00:12:21,130
‫That's impossible.

228
00:12:21,250 --> 00:12:23,470
‫So what this is does when you want to update.

229
00:12:23,470 --> 00:12:32,080
‫So what what it does is says, OK, let's just write the new value into a brand new page and then leave

230
00:12:32,080 --> 00:12:40,360
‫the old page as a stale market, as stale so future people can clean it up if they want to use a future

231
00:12:40,360 --> 00:12:40,960
‫requests.

232
00:12:41,530 --> 00:12:42,890
‫And here's another limitation.

233
00:12:42,890 --> 00:12:46,350
‫And this is the you cannot clean up one page at a time.

234
00:12:46,660 --> 00:12:48,640
‫You have to clean a whole block.

235
00:12:48,760 --> 00:12:50,290
‫And you can see some pictures here.

236
00:12:50,290 --> 00:12:52,150
‫A block, a block has many pages.

237
00:12:52,880 --> 00:12:55,180
‫And I think it's eight or something like that.

238
00:12:55,190 --> 00:12:58,700
‫If I come with with a payload, I want to update a page.

239
00:12:59,770 --> 00:13:02,560
‫I have a block or the three pages.

240
00:13:02,830 --> 00:13:04,000
‫I have four pages.

241
00:13:04,000 --> 00:13:05,650
‫Let's write them abcde.

242
00:13:05,740 --> 00:13:07,510
‫We just write them back right there.

243
00:13:07,690 --> 00:13:09,220
‫OK, and then let's follow that.

244
00:13:09,220 --> 00:13:10,120
‫And you can see the second page.

245
00:13:10,120 --> 00:13:17,500
‫I took this from Wikipedia resources right now and then let's follow it up with writing e f g h.

246
00:13:17,500 --> 00:13:17,950
‫Right.

247
00:13:18,100 --> 00:13:19,540
‫And then so that's four pages.

248
00:13:19,540 --> 00:13:27,430
‫But I also want to update pages A, B, C, D, so an update what it is to do.

249
00:13:27,580 --> 00:13:34,630
‫It marks those old ABCDE pages as stale and then write new four pages in the same block.

250
00:13:34,810 --> 00:13:36,220
‫So now this blog is filled.

251
00:13:37,750 --> 00:13:43,840
‫If you have a new rights that you want to come in and you want to use this block, you cannot use it

252
00:13:44,290 --> 00:13:46,630
‫to the ISS, the drivers.

253
00:13:47,320 --> 00:13:50,260
‫This is filled, this block is filled.

254
00:13:50,260 --> 00:13:51,550
‫You cannot use it at all.

255
00:13:52,270 --> 00:13:52,620
‫Right.

256
00:13:52,870 --> 00:13:54,730
‫And as a result is just filled.

257
00:13:54,760 --> 00:13:57,240
‫You cannot use that block.

258
00:13:57,580 --> 00:14:03,460
‫So there is this process that's called garbage collection and it runs asynchronously.

259
00:14:03,790 --> 00:14:05,890
‫It cleans up this kind of thing.

260
00:14:05,890 --> 00:14:09,360
‫So so this block has four pages that is in use then.

261
00:14:09,360 --> 00:14:13,810
‫And it's costing us a lot of precious page memory space.

262
00:14:13,810 --> 00:14:14,110
‫Right.

263
00:14:14,500 --> 00:14:20,860
‫So what we do is, hey, garbage collection, move all the active pages from this block into a brand

264
00:14:20,860 --> 00:14:28,540
‫new block and then just slam that block a block shot clean, flush the whole thing, clean it up so

265
00:14:28,540 --> 00:14:29,410
‫I can use it.

266
00:14:29,410 --> 00:14:31,680
‫I can start using all the free spaces.

267
00:14:31,960 --> 00:14:38,740
‫So now my pages have moved to the block, the second block, and they have now for around four spaces

268
00:14:38,740 --> 00:14:40,570
‫that I can use to write the stuff.

269
00:14:40,870 --> 00:14:44,980
‫That is why is is these do not like updates because of this thing.

270
00:14:44,980 --> 00:14:45,190
‫Right.

271
00:14:45,200 --> 00:14:52,750
‫You you're marking this pressure old pages as stale and then writing in the new locations every time,

272
00:14:53,170 --> 00:14:58,570
‫as opposed of new, fresh, nice sequential rights.

273
00:14:58,870 --> 00:14:59,320
‫Right.

274
00:14:59,560 --> 00:15:03,760
‫Every time you just using new pages, you never touching the old pages.

275
00:15:03,890 --> 00:15:09,250
‫So if you don't touch old pages, if you don't see old data, you're never going to mark these pages

276
00:15:09,250 --> 00:15:16,030
‫as stale and you're never going to have blocks with stale pages, which is the worst thing you can do

277
00:15:16,390 --> 00:15:21,400
‫if you have blocks with stale pages that have been updated, the garbage collection is going to kick

278
00:15:21,400 --> 00:15:27,130
‫in and the water level is another thing that's going to kick in, which is it's going to start flushing

279
00:15:27,130 --> 00:15:32,650
‫and moving in new pages to talk to a new blog and then flushing blocks.

280
00:15:32,950 --> 00:15:38,730
‫All this work, all this is this is going to do as a side effect of your original, right.

281
00:15:39,130 --> 00:15:47,110
‫So updates are the worst thing you can do is the SSD like SSD like is it is always like beautiful,

282
00:15:47,320 --> 00:15:48,040
‫just single.

283
00:15:48,040 --> 00:15:48,370
‫Right.

284
00:15:48,370 --> 00:15:52,160
‫So that's why Beta is right.

285
00:15:52,270 --> 00:16:02,200
‫You if you go to the databases, beta causes this kind of behavior over time on SSD, because even if

286
00:16:02,200 --> 00:16:07,360
‫you're doing only inserts, Beta is still need to get updated gear.

287
00:16:07,360 --> 00:16:15,190
‫Updating in place a big three pointer you're updating in place.

288
00:16:15,190 --> 00:16:16,570
‫Beta is locations.

289
00:16:16,780 --> 00:16:18,670
‫You're rebalancing the trees.

290
00:16:19,330 --> 00:16:24,370
‫And when you do that, B or B, B plus three, when you do an update, what do you do with this is do

291
00:16:24,370 --> 00:16:25,390
‫you go to that page?

292
00:16:25,390 --> 00:16:26,020
‫What exists?

293
00:16:26,170 --> 00:16:32,390
‫Slavitt as stale and then take that value, take the new value and write it to a new blog.

294
00:16:32,710 --> 00:16:37,390
‫So now the more you do this, the more you're going to have blocks with stale data.

295
00:16:37,390 --> 00:16:41,640
‫And when you have blocks with stale data garbage collection, you'll get a kick in this sky.

296
00:16:41,650 --> 00:16:45,850
‫Start moving your Prussia's block to some one place, which is an expensive process.

297
00:16:46,090 --> 00:16:48,040
‫It's going to happen again and again and again.

298
00:16:48,610 --> 00:16:50,050
‫Eventually you will get.

299
00:16:50,080 --> 00:16:59,250
‫They down very, very quickly because it's going to you're going to start using all your pages in unnecessarily,

300
00:16:59,280 --> 00:17:00,930
‫let's put it this way.

301
00:17:01,850 --> 00:17:06,770
‫And when you start using all the pages and as deep as this, these have this are going on with where

302
00:17:06,770 --> 00:17:13,310
‫leveling where they essentially says, OK, I have this page sitting for years, nobody touch this,

303
00:17:13,490 --> 00:17:16,340
‫and nobody written in that page.

304
00:17:16,340 --> 00:17:18,500
‫And maybe I forgot to mention that.

305
00:17:18,500 --> 00:17:24,260
‫But that page has only limited no amount of time to write, after which it just dies.

306
00:17:24,410 --> 00:17:26,150
‫That page location just dies.

307
00:17:26,180 --> 00:17:30,520
‫You cannot write to it maybe 2500 times or whatever.

308
00:17:30,530 --> 00:17:32,600
‫There is always this is this has a lifetime.

309
00:17:33,200 --> 00:17:33,550
‫All right.

310
00:17:33,800 --> 00:17:41,120
‫So there will be if you without leveling, there will be pages that is very busy where you are writing

311
00:17:41,120 --> 00:17:43,130
‫to it a lot, this physical location.

312
00:17:43,850 --> 00:17:47,380
‫And there will be pages that are never touched.

313
00:17:47,540 --> 00:17:48,890
‫So there will be unbalanced.

314
00:17:48,910 --> 00:17:54,110
‫So we're levelling as this is the concept where it just really bana's this thing says, OK, hey, this

315
00:17:54,110 --> 00:17:56,900
‫is a cold page from 1998.

316
00:17:56,900 --> 00:17:57,650
‫Nobody touched it.

317
00:17:57,840 --> 00:18:04,550
‫Let me move it to a new location and let me use that precious location to write new pages because you're

318
00:18:04,550 --> 00:18:06,610
‫doing all these updates eventually.

319
00:18:07,510 --> 00:18:07,900
‫All right.

320
00:18:08,180 --> 00:18:15,410
‫So as the the more you're doing that, the more we're going to kick in and to start using all this pages,

321
00:18:15,410 --> 00:18:20,500
‫all uses these in a balanced manner to update's always kills Orissa's.

322
00:18:20,510 --> 00:18:21,600
‫The fast, doesn't I?

323
00:18:21,620 --> 00:18:25,700
‫When I say that I'm talking years and years, obviously I'm not talking about bouncier.

324
00:18:25,700 --> 00:18:25,950
‫Right.

325
00:18:27,170 --> 00:18:28,100
‫So, yeah, that's why.

326
00:18:29,710 --> 00:18:37,210
‫People came in with a log structure ministry as an append only, so that we are having nice inserts

327
00:18:37,780 --> 00:18:45,970
‫tucked in together so my blocks don't have these stale pages and always have a six consecutive pages

328
00:18:45,970 --> 00:18:47,290
‫that are never updated.

329
00:18:47,410 --> 00:18:52,350
‫And as a result, SSD, these love this stuff is going only have pages that are never updated.

330
00:18:52,360 --> 00:18:54,130
‫As a result, there are clean.

331
00:18:54,840 --> 00:19:00,310
‫If you go back to old pages and start updating them as they have to move them to a new location, they

332
00:19:00,310 --> 00:19:05,890
‫have this block now all of a sudden have a stale page and all of the other pages are reused.

333
00:19:06,100 --> 00:19:10,900
‫So it has to move those pages to somewhere else and flush it or use it.

334
00:19:10,900 --> 00:19:13,200
‫And that's a lot of work for this August.

335
00:19:13,340 --> 00:19:18,390
‫So that was the right amplification from the application point of view, from the database point to

336
00:19:18,400 --> 00:19:26,890
‫and from this is the so you're single to do marketing the to do app as done if you're doing that single

337
00:19:26,890 --> 00:19:31,810
‫update that the front end developer, things that you're doing a single day, but you're actually touching

338
00:19:31,810 --> 00:19:35,350
‫all these tables and those tables kick in the database.

339
00:19:35,350 --> 00:19:35,680
‫Right.

340
00:19:35,680 --> 00:19:40,120
‫Amplification, which is table has like an all seven index and in this day will have three indexes and

341
00:19:40,120 --> 00:19:42,220
‫this day will have, I don't know, four, four indexes.

342
00:19:42,670 --> 00:19:48,640
‫Those if you're touching those obviously values with indexes, then those corresponding indexes have

343
00:19:48,640 --> 00:19:49,240
‫to be touch.

344
00:19:49,610 --> 00:19:53,860
‫And here's another one right in Postgres.

345
00:19:53,860 --> 00:20:00,280
‫There is this concept of toast tables where a string, if there is a long string, it never stored in

346
00:20:00,280 --> 00:20:00,820
‫line.

347
00:20:00,820 --> 00:20:01,090
‫Right.

348
00:20:01,090 --> 00:20:02,470
‫Is not another table.

349
00:20:02,470 --> 00:20:03,430
‫That's a toast table.

350
00:20:03,670 --> 00:20:05,080
‫So that table need to be updated.

351
00:20:05,080 --> 00:20:08,440
‫So the right amplification just triples.

352
00:20:08,440 --> 00:20:12,340
‫If it is just like order of magnitude, there's a lot of work.

353
00:20:12,670 --> 00:20:16,900
‫And then now all these rights, which some of them are updated, some of them are insert.

354
00:20:17,910 --> 00:20:24,000
‫It translates to SSD, right, amplification, ad, the physical disc level.

355
00:20:24,570 --> 00:20:32,430
‫So all these pages that you have to update and then move the new page, mark the old pages still move

356
00:20:32,450 --> 00:20:41,970
‫the create a new page and then have this is the point, the logical page pointer to the new page physical

357
00:20:41,970 --> 00:20:46,830
‫page, because that doesn't have the physical and the logical as it is as the concept.

358
00:20:47,110 --> 00:20:53,340
‫I know guys, it's a very complex topic, but I just wanted to make a video roar like that.

359
00:20:53,800 --> 00:20:56,670
‫It's a very, very important topic to understand.

360
00:20:56,670 --> 00:21:02,490
‫And then we're going to have to make a new and more more videos about this and maybe I'll go make a

361
00:21:02,490 --> 00:21:05,190
‫practical video discussing this in details.

362
00:21:05,790 --> 00:21:08,580
‫This was a right amplification.

363
00:21:09,300 --> 00:21:10,820
‫Just keep an eye on it.

364
00:21:11,340 --> 00:21:12,450
‫It's really critical.

365
00:21:14,350 --> 00:21:14,530
‫By.