1
00:00:00,330 --> 00:00:08,550
‫Roe versus Collum oriented databased storages are two style of bewitch databases used to store their

2
00:00:08,550 --> 00:00:13,130
‫tables on disk, each have pros and cons.

3
00:00:13,440 --> 00:00:14,650
‫Nothing is perfect.

4
00:00:14,670 --> 00:00:17,130
‫There are use cases for column oriented.

5
00:00:17,220 --> 00:00:21,110
‫There are use cases for ROE oriented databases.

6
00:00:22,020 --> 00:00:29,790
‫And this video, I want to discuss what is a role oriented database or also called role store database,

7
00:00:29,790 --> 00:00:31,920
‫because that's that's the store type.

8
00:00:31,920 --> 00:00:35,550
‫How is it stored on desk right now?

9
00:00:35,610 --> 00:00:41,940
‫Got to also describe the column oriented database or also known as Column Sterol.

10
00:00:42,270 --> 00:00:46,290
‫Another fancy name is called Menar Database because.

11
00:00:46,290 --> 00:00:50,700
‫Yeah, why not just confuse everybody, buy new by inventing new names.

12
00:00:51,030 --> 00:00:52,500
‫So it's all in the same thing.

13
00:00:52,950 --> 00:00:56,280
‫Just see how things are stored on this.

14
00:00:56,280 --> 00:00:58,200
‫We're going to talk about the differences too.

15
00:00:58,410 --> 00:01:04,740
‫And then finally going to discuss pros and cons of both.

16
00:01:05,110 --> 00:01:06,480
‫How about we jump into it, guys?

17
00:01:06,600 --> 00:01:07,020
‫Auriga.

18
00:01:07,020 --> 00:01:09,420
‫So let's say we have a beautiful table like this.

19
00:01:10,140 --> 00:01:15,930
‫It's an employee's table, doesn't only have eight roles that have thousands of thousands of rows,

20
00:01:16,560 --> 00:01:18,330
‫but for some lazy we only have these.

21
00:01:18,360 --> 00:01:19,290
‫These are the columns.

22
00:01:20,340 --> 00:01:26,520
‫Straightforward and the first I want to bring attention to this, this is the real I.D. This is how

23
00:01:26,910 --> 00:01:33,720
‫databases it's a unique identifier that identifies that it's different than the primary key or this

24
00:01:33,720 --> 00:01:40,260
‫I.D. It's just something that has most of the time most databases have this internal mechanism to identify

25
00:01:40,260 --> 00:01:40,820
‫this role.

26
00:01:41,520 --> 00:01:43,230
‫And I want you to pay attention to this.

27
00:01:43,350 --> 00:01:50,970
‫So if I'm going to submit a bunch of queries on this table, I want to do this without any indexes because

28
00:01:50,970 --> 00:01:53,970
‫indexes will just complicate the manner.

29
00:01:53,970 --> 00:01:56,990
‫And by the way, we think about column versus roll.

30
00:01:57,050 --> 00:02:00,780
‫Again, I don't I want to take them out of the equation and think about them.

31
00:02:01,080 --> 00:02:03,030
‫Just we don't have any indexes.

32
00:02:03,060 --> 00:02:07,860
‫Were just scanning the table to try to answer these following questions.

33
00:02:08,100 --> 00:02:12,240
‫So these are the following queries that I'm going to execute on our database.

34
00:02:12,360 --> 00:02:14,580
‫And then I'm going to ask you the same exact query.

35
00:02:14,580 --> 00:02:17,070
‫Don't call column database and see which one does best.

36
00:02:17,670 --> 00:02:22,590
‫They're for each style of how we're going to store these stables.

37
00:02:22,600 --> 00:02:22,970
‫Right.

38
00:02:22,980 --> 00:02:28,470
‫The same thing we're going to store tomorrow, the same day we're going to start on column storage and

39
00:02:28,470 --> 00:02:30,800
‫then we're going to look at how these queries perform.

40
00:02:30,810 --> 00:02:35,640
‫So select first name from Blois, where Social Security number 666, The Devil.

41
00:02:36,090 --> 00:02:41,310
‫And then we're going to select store from employee where ID equal one very simple game.

42
00:02:41,610 --> 00:02:43,200
‫And the final is to select some.

43
00:02:43,590 --> 00:02:46,560
‫We're just going to sum all the salaries of the employees for some reason.

44
00:02:48,720 --> 00:02:51,180
‫So the first one, rule oriented databases.

45
00:02:51,600 --> 00:02:58,440
‫So for me oriented databases, tables are stored as rolls on disk.

46
00:02:58,950 --> 00:03:02,280
‫And this might sound a little bit confusing when you first read it.

47
00:03:02,610 --> 00:03:07,260
‫Of course, Hosain, how else are you going to store stuff in the day you get a store roll by it?

48
00:03:07,260 --> 00:03:07,670
‫All right.

49
00:03:07,950 --> 00:03:15,600
‫So if you think of the disk as a blob, huge blob store of just things that you just just a bunch of

50
00:03:15,600 --> 00:03:22,250
‫bits and bytes, then the rows are just contiguous, one after the other.

51
00:03:22,260 --> 00:03:27,750
‫So the ID followed by the first name, followed by the second one last thing that followed by their

52
00:03:27,750 --> 00:03:31,500
‫search was to go down and then comma and then the second row starts.

53
00:03:31,500 --> 00:03:31,960
‫Right.

54
00:03:32,010 --> 00:03:33,030
‫And then so on.

55
00:03:33,030 --> 00:03:35,520
‫Each row obviously has a different size.

56
00:03:35,640 --> 00:03:38,030
‫It could be variable size because I don't know.

57
00:03:38,220 --> 00:03:42,570
‫So like my my name is John Smith and someone else named Hussein Mohamed.

58
00:03:43,110 --> 00:03:43,360
‫Right.

59
00:03:43,360 --> 00:03:44,460
‫So that's a longer name.

60
00:03:44,460 --> 00:03:47,040
‫So it's going to occupy more storage on disk, obviously.

61
00:03:47,190 --> 00:03:49,920
‫And that's the rules can get very much for simplicity.

62
00:03:49,920 --> 00:03:52,070
‫Just think of was their fixed size roles.

63
00:03:52,530 --> 00:03:52,910
‫All right.

64
00:03:53,220 --> 00:03:56,760
‫When I want to read a particular row.

65
00:03:58,030 --> 00:04:08,920
‫I go, I do an operation on desk called, I read when I read this stuff that controller, whether this

66
00:04:08,920 --> 00:04:16,230
‫is an SSD drive or a hard disk, reads and blocks, it doesn't read like bit by bit more.

67
00:04:16,240 --> 00:04:19,350
‫Oh, but zero one zero one doesn't really like that.

68
00:04:19,360 --> 00:04:20,160
‫It really just a block.

69
00:04:20,200 --> 00:04:28,030
‫So let's shove 512 bytes at a time and when you or I don't know, a thousand twenty four bytes at a

70
00:04:28,030 --> 00:04:29,920
‫time depends on the block size.

71
00:04:29,930 --> 00:04:31,270
‫There's something called block side.

72
00:04:31,570 --> 00:04:37,410
‫And once you read that you're going to get not only one roll, you could get one, two, three, four,

73
00:04:37,420 --> 00:04:39,550
‫five rows in a given block.

74
00:04:39,550 --> 00:04:42,340
‫So you get the act of doing one.

75
00:04:42,550 --> 00:04:48,970
‫You can give you five rows and just not all these five rows, five rows and all their columns or their

76
00:04:48,970 --> 00:04:49,440
‫values.

77
00:04:49,900 --> 00:04:50,200
‫So.

78
00:04:50,350 --> 00:04:53,290
‫So whether you want them or not, you're going to get everything.

79
00:04:53,320 --> 00:04:54,560
‫You're going to get all the columns.

80
00:04:54,610 --> 00:04:56,490
‫So think about that.

81
00:04:56,920 --> 00:05:02,450
‫So technically, you need more Io's to find the row you're looking for.

82
00:05:03,070 --> 00:05:08,020
‫So imagine you're doing us a table sequential scan on on the entire table.

83
00:05:08,320 --> 00:05:16,810
‫So if you're looking for employee with a Social Security number 666, you're going to have to do a block.

84
00:05:16,810 --> 00:05:18,760
‫I oh, you got, I don't know, seven rows.

85
00:05:19,060 --> 00:05:21,430
‫Then search these rows one by one.

86
00:05:21,430 --> 00:05:25,450
‫You're only interested in the Social Security number, so you're going to only pick that field.

87
00:05:25,610 --> 00:05:28,060
‫Unfortunately, you you got all the fields.

88
00:05:28,480 --> 00:05:29,770
‫Not saying that's bad.

89
00:05:29,770 --> 00:05:30,580
‫Could be bad.

90
00:05:30,580 --> 00:05:31,150
‫Could be good.

91
00:05:31,150 --> 00:05:34,750
‫Depends what you're doing, but that's what you have to work with.

92
00:05:34,960 --> 00:05:42,190
‫So you have to do a lot of Io's to to to actually get to the role that you were looking for because

93
00:05:42,190 --> 00:05:49,270
‫you spent a lot of wasteful reads pulling columns that you didn't need and you don't have a choice.

94
00:05:49,720 --> 00:05:49,970
‫Right.

95
00:05:49,990 --> 00:05:55,120
‫So that's that's just one thing you have to think about with your oriented as I'm in there.

96
00:05:55,120 --> 00:05:56,920
‫But no, they are beautiful.

97
00:05:56,920 --> 00:05:59,980
‫They are elegantly designed because they are so simple.

98
00:06:00,010 --> 00:06:00,850
‫They are just the rules.

99
00:06:01,180 --> 00:06:03,820
‫And you can easily think about them.

100
00:06:03,820 --> 00:06:07,750
‫You can easily implement them compared to what we're going to come up next.

101
00:06:08,200 --> 00:06:09,550
‫So let's dive deep into this.

102
00:06:09,580 --> 00:06:14,680
‫So we're going to take our table and then we're going to convert into a rule based on desk.

103
00:06:14,680 --> 00:06:16,570
‫How would how will this look like?

104
00:06:16,900 --> 00:06:17,620
‫Let's take a look.

105
00:06:18,730 --> 00:06:23,820
‫This is how we're going to look like, so imagine this gray box as a block, right?

106
00:06:24,070 --> 00:06:28,480
‫And for simplicity, I bought I put two rows in the same block.

107
00:06:28,870 --> 00:06:29,470
‫Could be three.

108
00:06:29,470 --> 00:06:30,010
‫Could be four.

109
00:06:30,010 --> 00:06:30,630
‫Could be five.

110
00:06:30,820 --> 00:06:31,620
‫Whoknows, right.

111
00:06:31,630 --> 00:06:32,710
‫Depends on the block size.

112
00:06:32,710 --> 00:06:43,150
‫But think about this as as a given block, each gray box takes one eye or jump to the controller to

113
00:06:43,150 --> 00:06:44,860
‫pull that block.

114
00:06:45,040 --> 00:06:47,740
‫So once you did you ask for this block.

115
00:06:47,950 --> 00:06:49,300
‫You got everything son.

116
00:06:49,390 --> 00:06:50,800
‫You got everything right.

117
00:06:51,040 --> 00:06:53,040
‫So now this is how it is stored.

118
00:06:53,050 --> 00:06:54,100
‫Basically the raw.

119
00:06:54,310 --> 00:06:55,270
‫Come on, really.

120
00:06:55,270 --> 00:06:55,630
‫Come on.

121
00:06:55,630 --> 00:06:58,600
‫It's just literally consequent right.

122
00:06:58,600 --> 00:06:59,020
‫Values.

123
00:06:59,020 --> 00:07:07,720
‫But I just add a comma for you guys to understand that ID first names like a last name SSN or the celery

124
00:07:08,050 --> 00:07:11,950
‫birth date, I guess when this John the occupation and so on.

125
00:07:11,980 --> 00:07:17,370
‫And so you can see that and this I added this so it can indicate the second drawer.

126
00:07:17,860 --> 00:07:19,660
‫It doesn't really exist.

127
00:07:19,660 --> 00:07:22,510
‫It's just an identifier like to allow that.

128
00:07:22,510 --> 00:07:27,820
‫Oh, this is actually a second row that just started and an arrow just ended.

129
00:07:27,820 --> 00:07:31,720
‫And you can see that we these are the two rows, two rows tourism.

130
00:07:32,710 --> 00:07:37,860
‫So let's execute our first query against a row oriented database.

131
00:07:38,710 --> 00:07:46,920
‫I'm looking for the first name for employee where their Social Security number 666, that four.

132
00:07:47,470 --> 00:07:47,890
‫All right.

133
00:07:48,310 --> 00:07:49,030
‫So let's pull it.

134
00:07:50,830 --> 00:07:51,410
‫What do we do?

135
00:07:52,000 --> 00:07:57,040
‫We ask because I don't know, I have to start from the top, right, because I don't know where it employs

136
00:07:57,040 --> 00:08:00,460
‫six six six is something like this.

137
00:08:00,760 --> 00:08:07,510
‫No, this field has any relation to the role itself is just a random value.

138
00:08:07,680 --> 00:08:08,010
‫Right.

139
00:08:08,290 --> 00:08:10,960
‫So I have to pull the first block of the table.

140
00:08:11,590 --> 00:08:14,230
‫Does sex, sex, sex exist in this NUPE?

141
00:08:15,340 --> 00:08:20,800
‫This is two to two, and this guy is one one one, that's the social worker, so let's pull the second

142
00:08:20,800 --> 00:08:22,050
‫block, right.

143
00:08:23,230 --> 00:08:24,430
‫Nope, it doesn't exist.

144
00:08:24,460 --> 00:08:26,370
‫This is four four four three three three, right.

145
00:08:26,650 --> 00:08:29,620
‫And yeah, it might look sorted, but it doesn't have to be sorted.

146
00:08:29,620 --> 00:08:34,300
‫Guys, just this is just I'm I'm walking through that area with you.

147
00:08:34,840 --> 00:08:35,070
‫Right.

148
00:08:35,080 --> 00:08:36,430
‫Again, noindex is nothing.

149
00:08:37,800 --> 00:08:38,760
‫Now, I'm going to.

150
00:08:39,740 --> 00:08:46,700
‫Pull the next block, so I did three reads and then I happen to find that six, six, six.

151
00:08:47,060 --> 00:08:49,890
‫So I found what I'm looking for now.

152
00:08:50,000 --> 00:08:52,570
‫I want the first name.

153
00:08:52,940 --> 00:09:00,470
‫That's what I don't need to do any extra read y because the first name is right there in the memory

154
00:09:00,470 --> 00:09:06,770
‫now, because once I pull this block, it's a memory, it's in the ramp and when I ask for it, I don't

155
00:09:06,770 --> 00:09:09,580
‫have to go back to the desk because yeah, guess what, I just pulled it.

156
00:09:09,710 --> 00:09:10,750
‫I found it.

157
00:09:10,760 --> 00:09:12,470
‫Now I just think this is the first thing.

158
00:09:12,480 --> 00:09:17,030
‫And I know because the position this position number two is the first time to pull it.

159
00:09:17,330 --> 00:09:25,700
‫So once I find something, asking for extra columns are really cheap because we we already got them

160
00:09:26,390 --> 00:09:27,500
‫right again.

161
00:09:27,510 --> 00:09:28,820
‫Guys, you have to think about this.

162
00:09:28,820 --> 00:09:34,400
‫Whatever I'm going to explain to you, there is no bad or good depends on your use case, depending

163
00:09:34,400 --> 00:09:36,110
‫on your query.

164
00:09:36,110 --> 00:09:37,670
‫That depends on your clauses.

165
00:09:37,940 --> 00:09:40,610
‫And based on that, you pick roll versus column this week.

166
00:09:40,700 --> 00:09:41,410
‫I'm going to come to that.

167
00:09:41,660 --> 00:09:42,720
‫So that's the first query.

168
00:09:43,760 --> 00:09:44,690
‫Let's do another query.

169
00:09:46,020 --> 00:09:46,960
‫Is there another way?

170
00:09:47,790 --> 00:09:50,100
‫No, I guess it's executed immediately.

171
00:09:50,520 --> 00:09:55,410
‫All right, so we're going to do a select star from employee where I'd equal one, right?

172
00:09:55,740 --> 00:10:03,630
‫And since it's already equal one, this if this is a sequence, the database can do tricks where it

173
00:10:03,630 --> 00:10:10,140
‫can link this sequence with that idea says, OK, I.D. one is on four thousand and one.

174
00:10:10,140 --> 00:10:16,410
‫So I know actually where to jump and pull this roll this block.

175
00:10:16,740 --> 00:10:17,040
‫Right.

176
00:10:17,130 --> 00:10:21,480
‫It doesn't have to read other things, but it happened to be that this the first book is what we're

177
00:10:21,480 --> 00:10:22,040
‫looking for.

178
00:10:22,050 --> 00:10:24,150
‫So yeah, we're lucky I guess.

179
00:10:24,510 --> 00:10:29,790
‫But even if we don't have this track where we link the ID to the real ID, we still have to go through

180
00:10:29,790 --> 00:10:30,480
‫each block.

181
00:10:30,480 --> 00:10:34,050
‫And once we find the block, we won't guess what.

182
00:10:34,680 --> 00:10:36,420
‫Now I want all the columns.

183
00:10:37,550 --> 00:10:38,240
‫Can I do this?

184
00:10:38,540 --> 00:10:39,320
‫Absolutely.

185
00:10:39,680 --> 00:10:48,530
‫That is so cheap because all the columns are already a memory, assuming they are not vertically partitioned

186
00:10:48,530 --> 00:10:50,600
‫into another location.

187
00:10:50,600 --> 00:10:52,280
‫Right, vertical partition.

188
00:10:52,310 --> 00:10:55,970
‫I talked about that and many times and in my channels and my courses.

189
00:10:56,330 --> 00:11:00,250
‫So vertical partitioning where we take a column and put it somewhere else.

190
00:11:00,260 --> 00:11:00,550
‫Right.

191
00:11:00,740 --> 00:11:03,970
‫This is not the case where everything is in one beautiful block.

192
00:11:03,980 --> 00:11:10,670
‫So I, I pulled the block, I asked for the I.D., I found the I.D. and now the user is asking for every

193
00:11:10,670 --> 00:11:11,020
‫field.

194
00:11:11,090 --> 00:11:15,350
‫I always tell people that, hey, selects those bad guys, stay away from it.

195
00:11:15,350 --> 00:11:18,080
‫But it really depends what you're trying to do.

196
00:11:18,080 --> 00:11:22,310
‫If you have indexes, try to do less of that stuff.

197
00:11:22,370 --> 00:11:27,260
‫So next thing, we're going to learn that it's the worst thing you can do in a column oriented database.

198
00:11:27,500 --> 00:11:29,020
‫So let's start enroll.

199
00:11:29,960 --> 00:11:31,400
‫You can live without it, I guess.

200
00:11:31,400 --> 00:11:31,880
‫Still.

201
00:11:32,960 --> 00:11:37,370
‫Just just be aware of that, so that's cheap, that's relatively cheap because I'm going to pull all

202
00:11:37,370 --> 00:11:42,080
‫the roads and everything and memory beautiful assume fit in memory.

203
00:11:42,080 --> 00:11:45,480
‫So could it could only be one block.

204
00:11:45,500 --> 00:11:50,500
‫Not necessarily because a road can span multiple blocks if it's too huge.

205
00:11:50,510 --> 00:11:57,920
‫I say we have a, I don't know, some blob here field and I it's a bad idea to store Blob's in line

206
00:11:57,920 --> 00:12:01,280
‫and you way, but let's assume you do.

207
00:12:01,280 --> 00:12:06,680
‫But that means this block, this road can span seven blocks for example.

208
00:12:06,860 --> 00:12:09,830
‫So you need to read all the block.

209
00:12:09,830 --> 00:12:10,460
‫Not necessarily.

210
00:12:10,460 --> 00:12:11,350
‫The data are smart.

211
00:12:11,540 --> 00:12:12,700
‫It's going to be the first block.

212
00:12:12,710 --> 00:12:15,020
‫Find the ID and only the rows.

213
00:12:15,020 --> 00:12:19,970
‫The, the columns you asked for is going to ask for the rest of the block to to fetch your stuff.

214
00:12:20,210 --> 00:12:21,850
‫Then I would have been doing this for years.

215
00:12:21,860 --> 00:12:23,120
‫They know what they doing.

216
00:12:23,120 --> 00:12:27,220
‫But I want you just I want to explain how is doing this stuff.

217
00:12:27,680 --> 00:12:28,090
‫Awesome.

218
00:12:29,060 --> 00:12:30,380
‫So let's do the same thing.

219
00:12:30,380 --> 00:12:33,530
‫But in aggregate this is called an aggregate function, which is an aggregate.

220
00:12:33,530 --> 00:12:36,080
‫It's just grouping by something.

221
00:12:36,080 --> 00:12:36,350
‫Right.

222
00:12:36,500 --> 00:12:42,410
‫So you work on one or a few columns, mainly one or less.

223
00:12:42,750 --> 00:12:45,440
‫How do you work with all of that's just

224
00:12:48,590 --> 00:12:49,140
‫an idea.

225
00:12:49,420 --> 00:12:53,520
‫OK, so you work with a few columns or let's say one in this example.

226
00:12:53,660 --> 00:12:56,960
‫OK, so select some salary from employers.

227
00:12:56,960 --> 00:12:57,740
‫Just some.

228
00:12:57,740 --> 00:12:58,400
‫Every other.

229
00:12:59,850 --> 00:13:00,510
‫Let's do this.

230
00:13:01,540 --> 00:13:02,840
‫So we pick the first block.

231
00:13:03,700 --> 00:13:06,790
‫Well, tough luck, I only wanted to say tough luck.

232
00:13:06,820 --> 00:13:08,350
‫I got everything again.

233
00:13:08,380 --> 00:13:10,050
‫No indexes here with indexes.

234
00:13:10,060 --> 00:13:15,820
‫This could be extremely fast, but I'm I'm just telling you what the database does without indexes.

235
00:13:16,060 --> 00:13:17,560
‫So you pull this oh.

236
00:13:17,710 --> 00:13:18,430
‫Back this early.

237
00:13:18,430 --> 00:13:21,710
‫It's one thousand and one and one thousand two.

238
00:13:21,730 --> 00:13:23,160
‫So we summed those two.

239
00:13:23,500 --> 00:13:26,520
‫Now let's pull the next block so we get the Mauro's.

240
00:13:26,710 --> 00:13:32,260
‫OK, one thousand one hundred and three one of Riquet went off OK, some of them.

241
00:13:32,470 --> 00:13:36,330
‫And then pull one oh five or six, some of them.

242
00:13:36,910 --> 00:13:45,220
‫So guys, every block we try, we pull, we're pulling Roy these are pulling first names and last names,

243
00:13:45,220 --> 00:13:49,050
‫bonuses and we're pulling salary opening there, which we never use.

244
00:13:49,540 --> 00:13:49,890
‫Right.

245
00:13:50,020 --> 00:13:57,280
‫So imagine if I only asked for the for the salaries and if they were they were grouped nicely, that

246
00:13:57,280 --> 00:13:58,560
‫would be awesome, wouldn't it?

247
00:13:59,080 --> 00:14:03,820
‫But now, unfortunately, we're asking for salary, but we're pulling all the rules.

248
00:14:03,820 --> 00:14:05,360
‫So that's a lot of iio.

249
00:14:05,410 --> 00:14:07,930
‫So again, this is not just one read.

250
00:14:08,380 --> 00:14:09,940
‫It can appear in this.

251
00:14:10,210 --> 00:14:10,930
‫He has a one read.

252
00:14:11,110 --> 00:14:14,110
‫If the road is long, you can read seven blocks.

253
00:14:14,110 --> 00:14:15,880
‫So that's seven Io's, right?

254
00:14:16,120 --> 00:14:22,960
‫I mean, the other databases can do a trick where they send one request asynchronously, read multiple

255
00:14:22,960 --> 00:14:23,410
‫blocks.

256
00:14:23,410 --> 00:14:31,660
‫But regardless, you're hitting the desk, which which every time you hit the desk is a duck dies essentially.

257
00:14:31,690 --> 00:14:31,960
‫Right.

258
00:14:31,960 --> 00:14:34,930
‫So try to save Ducks' as much as possible.

259
00:14:34,930 --> 00:14:36,060
‫Save the ducks, guys.

260
00:14:36,130 --> 00:14:36,820
‫Save the ducks.

261
00:14:37,330 --> 00:14:39,160
‫So yeah.

262
00:14:39,340 --> 00:14:42,370
‫Reading, reading and well something well something all of that stuff.

263
00:14:42,970 --> 00:14:45,190
‫That's not so bad I guess.

264
00:14:45,190 --> 00:14:45,400
‫Yeah.

265
00:14:45,790 --> 00:14:48,370
‫But you read unnecessary information.

266
00:14:48,370 --> 00:14:51,550
‫That means you almost read the entire table.

267
00:14:53,080 --> 00:14:55,180
‫But used very.

268
00:14:56,780 --> 00:15:03,190
‫Very small portion of it that's extremely inefficient, so raw data doesn't do well, this kind of squares,

269
00:15:03,190 --> 00:15:08,600
‫if you think about it, it doesn't mean it's slow, really depends what you mean by slow.

270
00:15:08,890 --> 00:15:14,590
‫I mean, you you have run these kind of databases, do all kind of tricks to be fast.

271
00:15:14,590 --> 00:15:20,770
‫But if you think about the logic and I'm not saying this is how databases absolutely work, they have

272
00:15:20,770 --> 00:15:21,640
‫all sorts of trick.

273
00:15:21,640 --> 00:15:23,010
‫Again, multi threading.

274
00:15:23,110 --> 00:15:27,460
‫They send multiple threads, Boscastle and multiple thread to read asynchronously.

275
00:15:27,460 --> 00:15:30,490
‫So so they do all sorts of work to do fast.

276
00:15:30,820 --> 00:15:34,750
‫But let's discuss how column oriented databases work.

277
00:15:35,530 --> 00:15:36,910
‫So old Coolmore into databases.

278
00:15:36,910 --> 00:15:41,480
‫If you think about them, tables are stored as columns first and disk.

279
00:15:41,500 --> 00:15:43,090
‫So think about.

280
00:15:44,070 --> 00:15:45,270
‫The first name called.

281
00:15:46,670 --> 00:15:48,670
‫The I'd call the last name call.

282
00:15:48,830 --> 00:15:53,450
‫So what they do is they take the first name and they take all possible values.

283
00:15:54,170 --> 00:15:56,240
‫John, Melissa.

284
00:15:58,600 --> 00:16:08,080
‫Rik, Paul, Hussein, everybody, and pull all of these into and in and save them and desk as consequence

285
00:16:08,080 --> 00:16:13,990
‫value, and then once that table is done and that their last value of the first name is done, they

286
00:16:13,990 --> 00:16:17,830
‫start with the second one last name and then and then so on.

287
00:16:18,250 --> 00:16:20,740
‫So you might say, oh, why are we doing this?

288
00:16:20,740 --> 00:16:21,710
‫And we're going to come to that.

289
00:16:22,390 --> 00:16:31,740
‫So a single book I'll read to the table features multiple columns with all matching rows.

290
00:16:32,290 --> 00:16:35,080
‫So it features a column.

291
00:16:35,990 --> 00:16:42,920
‫It could be fish one column or multiple depends, it depends like if you have a small table, you can

292
00:16:42,920 --> 00:16:43,790
‫fish multiple one.

293
00:16:43,880 --> 00:16:46,250
‫If you ask the database's, hey, give me this.

294
00:16:46,430 --> 00:16:48,110
‫You're going to get a lot of arrows.

295
00:16:48,710 --> 00:16:50,140
‫That's the trick here.

296
00:16:50,450 --> 00:16:53,660
‫So single call with a lot of those lists.

297
00:16:53,660 --> 00:16:58,910
‫Io's are required to get more values of a given column.

298
00:16:59,630 --> 00:17:00,530
‫Right, because.

299
00:17:01,600 --> 00:17:09,520
‫If you want to get, as we said, the first hundred salaries, that is extremely efficient and goal

300
00:17:09,520 --> 00:17:15,970
‫oriented because they just go to the place where we know we started the salary column and just walk

301
00:17:15,970 --> 00:17:18,960
‫through there because you get to see the values just consequent.

302
00:17:19,000 --> 00:17:23,620
‫We're going to go to the example and they are great on online and process, as we're going to see and

303
00:17:23,620 --> 00:17:24,910
‫then in the coming slides.

304
00:17:25,540 --> 00:17:25,840
‫All right.

305
00:17:25,840 --> 00:17:29,370
‫So let's take the same exact table and start as Comb's.

306
00:17:29,380 --> 00:17:30,250
‫Here's how it looks like.

307
00:17:30,250 --> 00:17:32,290
‫Again, the ride is very critical here.

308
00:17:32,800 --> 00:17:34,270
‫The I.D. field is this.

309
00:17:34,450 --> 00:17:36,430
‫This is the first name that can last name.

310
00:17:36,430 --> 00:17:39,760
‫I keep saying second Social Security number.

311
00:17:40,060 --> 00:17:44,440
‫This is the salary, the date of birth, the that title, I think.

312
00:17:44,440 --> 00:17:45,610
‫And this is the joint date.

313
00:17:46,030 --> 00:17:46,720
‫Look at this.

314
00:17:47,050 --> 00:17:48,490
‫So look at this.

315
00:17:48,770 --> 00:17:53,230
‫Want the ID one and this, which is this thousand and one.

316
00:17:53,590 --> 00:17:57,400
‫I need to withdraw three thousand three.

317
00:17:57,520 --> 00:17:59,650
‫Why we're going to we're going to see look at this.

318
00:17:59,830 --> 00:18:02,140
‫The wrong idea is duplicated in every column.

319
00:18:03,620 --> 00:18:10,310
‫So just like that, what is the first thing that comes to mind or anything is going to be painful?

320
00:18:11,390 --> 00:18:19,510
‫Because anything if you delete roll 2004, you have to boo boo, boo, boo, boo boo.

321
00:18:20,630 --> 00:18:25,550
‫You have to go and mark all these stupid columns or these blocks.

322
00:18:25,560 --> 00:18:26,780
‫Remember, these are blocks, right?

323
00:18:27,140 --> 00:18:33,680
‫So and for fun, I split some of them and multiple blocks doesn't mean they are like, I know this is

324
00:18:33,680 --> 00:18:35,690
‫an integer, for example, or string.

325
00:18:36,050 --> 00:18:43,910
‫I just split them for me that OK, we couldn't store all all four rolls here in one block, so we had

326
00:18:43,910 --> 00:18:45,320
‫to split it another block.

327
00:18:45,320 --> 00:18:45,580
‫Right.

328
00:18:45,590 --> 00:18:47,660
‫So I can show you a different kind of examples.

329
00:18:48,110 --> 00:18:49,850
‫Again, take it with a grain of salt.

330
00:18:49,850 --> 00:18:51,790
‫I just want to explain this things to you.

331
00:18:51,920 --> 00:18:57,530
‫So this is one block to block to block two blocks, one block, one block, one block, one block.

332
00:18:57,540 --> 00:18:58,460
‫OK, got it.

333
00:18:59,600 --> 00:19:00,700
‫So this is how it works.

334
00:19:00,860 --> 00:19:02,810
‫So they just store all this stuff in columns.

335
00:19:02,810 --> 00:19:08,150
‫So all the Eidsvold the entire the entire table's here.

336
00:19:08,300 --> 00:19:08,600
‫Right.

337
00:19:08,750 --> 00:19:09,980
‫All the possible values.

338
00:19:10,280 --> 00:19:13,190
‫So this is not just one block.

339
00:19:13,190 --> 00:19:17,510
‫Could be hundreds of thousands of blocks if you have a lot of rows.

340
00:19:18,170 --> 00:19:20,270
‫Remember, everything need to be updated.

341
00:19:20,450 --> 00:19:26,480
‫If you add a new role, you have to get that updated in all of these logical structures.

342
00:19:26,840 --> 00:19:31,310
‫So what that being said, now we know how this works.

343
00:19:32,180 --> 00:19:33,860
‫Let's put it to the test.

344
00:19:33,980 --> 00:19:38,170
‫Select first name from employee where Social Security number six six six.

345
00:19:38,360 --> 00:19:43,220
‫So now the database will say, wait a second, you're looking for Social Security number.

346
00:19:43,580 --> 00:19:50,810
‫I don't need to look at any of other logical structures except for the Social Security, which is this

347
00:19:50,810 --> 00:19:51,170
‫puppy.

348
00:19:51,320 --> 00:19:55,730
‫This is the Social Security data descried.

349
00:19:55,730 --> 00:19:58,280
‫This is how I know this is the location.

350
00:19:58,580 --> 00:20:03,890
‫So in DESC, I know where to point my needle and read this stuff.

351
00:20:04,310 --> 00:20:10,010
‫So to search for Social Security number, I only need to pull this right.

352
00:20:10,130 --> 00:20:12,590
‫So now I start reading block by block.

353
00:20:12,770 --> 00:20:16,970
‫I need first block and I got a lot of beautiful values.

354
00:20:16,970 --> 00:20:19,550
‫I got five five five four four, four, three, two, one, two, one.

355
00:20:21,230 --> 00:20:22,670
‫We did not get six six.

356
00:20:23,300 --> 00:20:23,960
‫So no.

357
00:20:24,020 --> 00:20:24,860
‫OK, no problem.

358
00:20:24,860 --> 00:20:26,810
‫Let's pull the second block.

359
00:20:26,960 --> 00:20:30,530
‫This is almost like the roll based, but now add the column level.

360
00:20:30,530 --> 00:20:31,220
‫Does that make sense.

361
00:20:31,970 --> 00:20:33,500
‫So now we got it.

362
00:20:33,860 --> 00:20:34,610
‫We found it.

363
00:20:36,260 --> 00:20:38,210
‫But now I need the first name.

364
00:20:39,170 --> 00:20:41,480
‫What first name is not here, son?

365
00:20:42,530 --> 00:20:47,230
‫You only got the six six now, you found that this is actually 2006.

366
00:20:47,450 --> 00:20:50,390
‫That's almost like an in how indexing works.

367
00:20:50,390 --> 00:20:53,480
‫If you think about it, this is how Posterous stores indexes.

368
00:20:53,660 --> 00:20:57,680
‫Actually, if you think about this as an index, it's almost very similar.

369
00:20:57,870 --> 00:20:58,220
‫Right?

370
00:20:58,610 --> 00:21:01,760
‫It just points to the row now, 2006.

371
00:21:02,730 --> 00:21:11,190
‫Roll twenty six, I know this value, I know the roll I'm looking for, this is very critical and I

372
00:21:11,250 --> 00:21:13,980
‫am asked for the first name was the first name.

373
00:21:13,980 --> 00:21:15,540
‫The first name is right here.

374
00:21:15,540 --> 00:21:16,680
‫This is the first name.

375
00:21:16,680 --> 00:21:19,620
‫So I'm not going to read these, blah.

376
00:21:19,860 --> 00:21:21,660
‫I'm only going to jump right here.

377
00:21:22,110 --> 00:21:22,440
‫Right.

378
00:21:22,570 --> 00:21:26,220
‫So obviously we found this row, as we said, and then we can jump.

379
00:21:26,220 --> 00:21:28,980
‫Right, not the first block.

380
00:21:28,980 --> 00:21:31,290
‫I'm going to jump to this block, you might say.

381
00:21:31,290 --> 00:21:35,940
‫I was saying, how did you know that it's in this block and you didn't pull this block because I know

382
00:21:35,940 --> 00:21:36,600
‫the wrong number.

383
00:21:36,990 --> 00:21:44,490
‫And the database does all sorts of trick to to say, OK, this block has only rose from one thousand

384
00:21:44,490 --> 00:21:45,900
‫and one to the 1004.

385
00:21:45,930 --> 00:21:46,280
‫Right.

386
00:21:46,560 --> 00:21:54,240
‫So I'm going to jump to block number seven hundred and three because it exactly locate the row I want

387
00:21:54,540 --> 00:21:58,020
‫because they have all this metadata mumbo-jumbo.

388
00:21:58,020 --> 00:21:58,410
‫Right.

389
00:21:58,860 --> 00:22:00,750
‫Again, that's back to the our question.

390
00:22:00,960 --> 00:22:04,740
‫The 1006 link with a block is all there.

391
00:22:04,770 --> 00:22:07,380
‫They know and they pull it and they find it.

392
00:22:07,380 --> 00:22:09,330
‫So we had to do one jump.

393
00:22:09,630 --> 00:22:10,400
‫One block, right.

394
00:22:10,650 --> 00:22:15,140
‫Second block read, jumped back to another block read and then we got it.

395
00:22:15,150 --> 00:22:15,600
‫So three.

396
00:22:16,080 --> 00:22:17,190
‫No, that's so bad.

397
00:22:17,190 --> 00:22:21,930
‫I guess let's just start from employee where it equal.

398
00:22:21,930 --> 00:22:28,410
‫OK, let's see how how come how our beautiful co index Culham Storage do.

399
00:22:28,470 --> 00:22:28,890
‫All right.

400
00:22:29,190 --> 00:22:30,300
‫So I'd equal one again.

401
00:22:30,300 --> 00:22:35,040
‫I don't have any knowledge of this so I'm going to start idy whereas idee this is dite.

402
00:22:35,040 --> 00:22:38,700
‫So I need only to read these structures.

403
00:22:38,700 --> 00:22:40,020
‫Right, let's put it.

404
00:22:41,480 --> 00:22:49,490
‫Found one thousand and one, so now I have knowledge of the role, I know which book to read.

405
00:22:50,480 --> 00:22:51,260
‫But.

406
00:22:53,320 --> 00:23:02,920
‫The user asked me for everything, OK, let's jump on, OK, we found that jump onto the first name.

407
00:23:03,930 --> 00:23:09,480
‫We know which blog to read, so this one put, that's one read, let's jump one over AIO.

408
00:23:10,510 --> 00:23:15,250
‫Last name, because they want the last name, they want everything read, OK, founded because I know

409
00:23:15,760 --> 00:23:22,300
‫I don't need to read this because I only know which block exactly in each column to read to find my

410
00:23:22,300 --> 00:23:22,690
‫value.

411
00:23:22,960 --> 00:23:23,310
‫All right.

412
00:23:23,970 --> 00:23:26,140
‫This is in fine read.

413
00:23:27,980 --> 00:23:37,200
‫Salary read, birthday read engineer, I have a bug here, I should have I didn't stretch it enough.

414
00:23:37,700 --> 00:23:38,780
‫Engineer Reid.

415
00:23:40,020 --> 00:23:41,340
‫And joined it read.

416
00:23:44,550 --> 00:23:45,520
‫What have we done, guys?

417
00:23:46,480 --> 00:23:55,390
‫This is the worst query, there is so much thrashing happening on disk that this is the worst thing

418
00:23:55,390 --> 00:23:55,880
‫you can do.

419
00:23:56,350 --> 00:24:03,160
‫Call them databases, just tanks when you do this, especially if you have a lot of columns and you're

420
00:24:03,160 --> 00:24:09,550
‫asking for everything, and especially if you have like a lot of and queries or queries here where you

421
00:24:09,550 --> 00:24:12,550
‫had to jump to find that in multiple rows.

422
00:24:14,240 --> 00:24:22,850
‫There will dude, dude, dude, that dude endou that I'm talking to both of you, dude, dude that do

423
00:24:22,850 --> 00:24:23,540
‫not do this.

424
00:24:24,020 --> 00:24:26,710
‫So let's start with the raw stories and columns.

425
00:24:26,720 --> 00:24:28,210
‫Stories do not do select stuff.

426
00:24:28,730 --> 00:24:34,790
‫This is do you know how many ducks die when you do this kind of queries.

427
00:24:34,790 --> 00:24:40,010
‫Guys, you go to the database, you read all these damn ducks, every IO kills a duck.

428
00:24:40,250 --> 00:24:40,580
‫Right.

429
00:24:40,580 --> 00:24:41,950
‫Save the ducks please, guys.

430
00:24:43,280 --> 00:24:46,190
‫All right, let's do this like some celery for our employees.

431
00:24:46,190 --> 00:24:49,700
‫Know that we saw how horrible column oriented databases can be.

432
00:24:49,940 --> 00:24:51,620
‫Let's see how great they are.

433
00:24:53,120 --> 00:24:55,940
‫I want to sum all the salaries.

434
00:24:55,940 --> 00:24:56,630
‫Yes, sir.

435
00:24:57,050 --> 00:24:58,100
‫What is the salary field?

436
00:24:58,370 --> 00:24:59,090
‫There it is.

437
00:24:59,900 --> 00:25:05,930
‫One read dun dun.

438
00:25:06,680 --> 00:25:10,130
‫Well, if you have if you live in multiple blocks, are you going to read the multiple blocks.

439
00:25:10,130 --> 00:25:11,210
‫But done.

440
00:25:11,690 --> 00:25:12,470
‫That's it.

441
00:25:13,070 --> 00:25:13,880
‫That's it.

442
00:25:14,510 --> 00:25:17,150
‫And if you have caching and stuff like that.

443
00:25:17,390 --> 00:25:17,960
‫Hmm.

444
00:25:18,980 --> 00:25:21,800
‫And by the way, I didn't mention something.

445
00:25:21,950 --> 00:25:27,770
‫If you, if people, multiple people have the same salary, column oriented databases do something called

446
00:25:27,770 --> 00:25:37,730
‫compression and aggregate, they actually if let's say if if three people have 100000 salary, OK,

447
00:25:38,690 --> 00:25:40,190
‫all of them have a lot of calories.

448
00:25:40,190 --> 00:25:40,420
‫Right.

449
00:25:40,640 --> 00:25:42,080
‫They have a lot of calories.

450
00:25:42,170 --> 00:25:49,010
‫And then you can you can just have one in three, one hundred thousand with a comma besides one thousand

451
00:25:49,010 --> 00:25:51,200
‫three, one thousand four, one thousand and five.

452
00:25:51,200 --> 00:25:52,300
‫All of them have the same one.

453
00:25:52,310 --> 00:25:55,490
‫So this is even more compact.

454
00:25:55,490 --> 00:26:00,590
‫So this will give me a lot of bang for my buck by just doing one block again.

455
00:26:00,590 --> 00:26:06,170
‫I sometimes if I if there is like a duplication new, this is the best thing for going the database

456
00:26:06,170 --> 00:26:10,520
‫that you just shove it in the same row because they go they were going to do it as an array.

457
00:26:10,520 --> 00:26:14,330
‫We talked about that a little bit in my Postgres Thirteen Polska thirteen.

458
00:26:14,660 --> 00:26:21,080
‫They just started doing this index deduplication I believe they call it B three D duplication.

459
00:26:21,080 --> 00:26:22,550
‫Postgres version thirteen.

460
00:26:22,730 --> 00:26:31,580
‫They start shoving more duplicated values in a single leaf node so they can compress comparison.

461
00:26:32,540 --> 00:26:35,000
‫So yeah, pros and cons.

462
00:26:35,480 --> 00:26:36,620
‫Is this thing perfect.

463
00:26:36,740 --> 00:26:38,240
‫Nope, nope, nope, nope.

464
00:26:38,240 --> 00:26:38,480
‫Nothing.

465
00:26:38,490 --> 00:26:40,260
‫But it depends on your use case guys.

466
00:26:40,580 --> 00:26:44,990
‫And if you ask me, I still prefer Rabassa to be honest.

467
00:26:45,020 --> 00:26:46,970
‫Just simplicity overrides and reads.

468
00:26:47,360 --> 00:26:52,250
‫But there are some cases where you need column orienteers to summarize.

469
00:26:52,250 --> 00:26:53,930
‫Right, right.

470
00:26:54,770 --> 00:26:57,770
‫And reads you want a simple structure.

471
00:26:57,770 --> 00:27:01,610
‫And we saw how complex column stores are.

472
00:27:01,730 --> 00:27:02,050
‫Right.

473
00:27:02,840 --> 00:27:03,790
‫They're very complex.

474
00:27:03,800 --> 00:27:04,970
‫They you have to duplicate that.

475
00:27:04,970 --> 00:27:09,690
‫You have to show that the the row IDs everywhere.

476
00:27:10,400 --> 00:27:13,430
‫That can be a little bit messy for right and slow.

477
00:27:13,480 --> 00:27:13,850
‫Right.

478
00:27:13,850 --> 00:27:19,340
‫And that's why a lot of people use column based and warehouses and stuff like that and lakes, data

479
00:27:19,340 --> 00:27:22,310
‫lakes where things are just almost static.

480
00:27:22,310 --> 00:27:26,720
‫They don't edit and they just do a lot of analytics on a single column.

481
00:27:26,720 --> 00:27:28,820
‫So that's powerful stuff.

482
00:27:28,820 --> 00:27:29,210
‫Right.

483
00:27:29,570 --> 00:27:32,540
‫So let's go through these pros and cons in general.

484
00:27:32,810 --> 00:27:37,430
‫So for based, it's very optimal for read and write, right.

485
00:27:37,430 --> 00:27:37,850
‫And end.

486
00:27:38,210 --> 00:27:39,290
‫Well, I'm going to say read and write.

487
00:27:39,290 --> 00:27:44,960
‫I'm going to say transactions in general or online transactional processing are they are very good for

488
00:27:44,960 --> 00:27:49,340
‫this because they are very simple in their implementation.

489
00:27:50,320 --> 00:27:56,650
‫So if something simple, we can enhance it and make it more and more efficient as we go, right, that's

490
00:27:56,650 --> 00:27:59,990
‫why a writer fast reads almost fast.

491
00:27:59,990 --> 00:28:03,060
‫It depends what how your career it looks like we saw some examples, right?

492
00:28:03,400 --> 00:28:11,410
‫Online transfer transfer, online transaction protocols and processing transactions are great for Rule-based

493
00:28:11,410 --> 00:28:16,510
‫because you start a transaction when we know what exactly what rules, what blocks to touch, we can

494
00:28:16,720 --> 00:28:23,800
‫write the wall, the right ahead log efficiently because we know exactly where we're changing right

495
00:28:24,100 --> 00:28:26,290
‫versus column, the small little bit.

496
00:28:26,290 --> 00:28:27,880
‫The structure is all over.

497
00:28:27,880 --> 00:28:34,330
‫The place that we have to do a scattershot compression isn't really as effective, though, because

498
00:28:34,330 --> 00:28:34,630
‫that.

499
00:28:36,450 --> 00:28:36,990
‫Think about it.

500
00:28:37,100 --> 00:28:40,030
‫No, I'm not talking about the do duplication.

501
00:28:40,080 --> 00:28:41,780
‫That's something you can do in the next.

502
00:28:41,830 --> 00:28:50,910
‫I'm talking about just the idea of having the values in the in the role.

503
00:28:52,020 --> 00:28:55,210
‫The role itself is almost distinct, right?

504
00:28:55,590 --> 00:28:59,520
‫It's almost like a hash because it consists of different fields.

505
00:28:59,620 --> 00:29:03,300
‫And so you can really easily compress it.

506
00:29:03,360 --> 00:29:03,750
‫Right.

507
00:29:04,410 --> 00:29:11,630
‫The compression algorithms is not going to find a lot of tweaks to compress because the values are not

508
00:29:12,120 --> 00:29:12,980
‫consequence.

509
00:29:13,020 --> 00:29:13,220
‫Right.

510
00:29:13,290 --> 00:29:16,200
‫Compared to the column oriented.

511
00:29:16,200 --> 00:29:17,390
‫Where are they?

512
00:29:17,460 --> 00:29:22,080
‫The consequent values are almost the same type and almost the same thing.

513
00:29:22,080 --> 00:29:22,390
‫Right.

514
00:29:22,590 --> 00:29:28,980
‫So you're going to get a lot of chances were of duplication, like you get a lot of people named John.

515
00:29:29,010 --> 00:29:29,390
‫Right.

516
00:29:29,640 --> 00:29:33,470
‫So John's are going to be squashed together.

517
00:29:33,630 --> 00:29:41,160
‫So compression is not as efficient withdraws because you get all these different things that are next

518
00:29:41,160 --> 00:29:42,630
‫to each other and are completely different.

519
00:29:42,640 --> 00:29:45,240
‫So we can't compress as effectively aggregation.

520
00:29:45,240 --> 00:29:51,020
‫We saw how poorly the Rome based database did with the with the aggregation.

521
00:29:51,300 --> 00:29:55,620
‫I mean, we had to query the whole table, but we we pulled.

522
00:29:56,490 --> 00:30:02,820
‫Lots of information that we didn't pick, so column oriented databases were more efficient when it comes

523
00:30:02,820 --> 00:30:08,920
‫to aggregation because we fetch only what we need and we we work on that.

524
00:30:08,940 --> 00:30:17,130
‫So if you if you started switching information or data that that are only what you need, then you're

525
00:30:17,130 --> 00:30:23,880
‫going to start you you'd be very, very efficient in your query and you only fetch the information that

526
00:30:23,880 --> 00:30:24,270
‫you need.

527
00:30:24,570 --> 00:30:24,800
‫Right.

528
00:30:25,050 --> 00:30:31,170
‫So obviously you're going to have a lot and you're going to finish faster than than raw oriented databases,

529
00:30:31,530 --> 00:30:33,480
‫efficient queries with multi columns.

530
00:30:33,480 --> 00:30:33,730
‫Right.

531
00:30:33,990 --> 00:30:40,680
‫So if you're picking multiple columns in a query or working with multiple columns, usually.

532
00:30:42,780 --> 00:30:44,130
‫Rural oriented or better?

533
00:30:44,380 --> 00:30:49,720
‫Right, and again, depends on talking only about the queries itself, right.

534
00:30:50,600 --> 00:30:56,420
‫So you can have a lot of columns if you want to, but if you're only working with a few of them, that's

535
00:30:56,660 --> 00:30:59,600
‫absolutely fine with column oriented databases, but.

536
00:31:01,130 --> 00:31:09,860
‫If you're actually working on touching all this Combes, I could be about right with Kamaran, with

537
00:31:10,010 --> 00:31:16,880
‫Wetherall, we saw it write a fetch on the road, give us almost all the columns against you unless

538
00:31:16,880 --> 00:31:18,190
‫you have vertical partition.

539
00:31:18,680 --> 00:31:23,870
‫So we'll call them based are slower because we have to update all these beautiful structures.

540
00:31:23,870 --> 00:31:26,570
‫Every column has almost a structure on itself.

541
00:31:26,600 --> 00:31:27,440
‫Think of it this way.

542
00:31:27,440 --> 00:31:28,360
‫How is this stored?

543
00:31:28,580 --> 00:31:31,880
‫So I need to know where is it located and touch it, touch it, touch it.

544
00:31:31,880 --> 00:31:32,690
‫Every column.

545
00:31:32,720 --> 00:31:39,260
‫Right, like very similar to indexes if you think they are perfect for online analytical processing.

546
00:31:39,260 --> 00:31:43,820
‫So if you're doing analysis, you're not touching, you're not writing perfect.

547
00:31:43,820 --> 00:31:45,010
‫Cormoran to databased.

548
00:31:45,650 --> 00:31:51,830
‫It compress greatly because all similar sampling are together.

549
00:31:52,070 --> 00:31:58,220
‫So you can put the compression algorithms can do magic on these things and compress them.

550
00:31:58,700 --> 00:32:03,500
‫Amazing for aggregation we so be done.

551
00:32:03,930 --> 00:32:04,210
‫Right.

552
00:32:04,220 --> 00:32:05,860
‫That was, that was amazing.

553
00:32:05,870 --> 00:32:11,320
‫I'm going to, I got, I got tired of it and finally it's inefficient for queries.

554
00:32:11,480 --> 00:32:16,580
‫Well the multiple comes if you're working, if you're asking for all columns, don't do a calm, don't

555
00:32:16,580 --> 00:32:17,060
‫do a column.

556
00:32:17,310 --> 00:32:22,550
‫I know there's a little bit of where you think that column oriented database's or Colmar into deadest

557
00:32:22,550 --> 00:32:26,540
‫or column number will be great if you're asking for a lot of columns.

558
00:32:26,540 --> 00:32:28,700
‫Nope, they are the worst of this stuff.

559
00:32:29,150 --> 00:32:37,160
‫I know guys Postgres, my school, other database's, most of them are Robair storage.

560
00:32:37,310 --> 00:32:45,620
‫However, S.P.C.A. and others have Oracle, I believe most databases by default.

561
00:32:45,620 --> 00:32:48,910
‫Gaudreau row row based storage.

562
00:32:49,700 --> 00:32:57,020
‫However, if you think about it, they have database storage engine and we talked about database storage

563
00:32:57,260 --> 00:33:01,160
‫and my database scores and my YouTube channel as well.

564
00:33:01,160 --> 00:33:07,400
‫I talked about that, how our database engine, you can swizzle this database engine for a given table

565
00:33:07,610 --> 00:33:15,230
‫so every table you can pick a table and you store this table as Collum store and you can pick another

566
00:33:15,230 --> 00:33:16,730
‫table and start as store.

567
00:33:17,240 --> 00:33:19,430
‫You can Switzler database engine.

568
00:33:19,610 --> 00:33:19,940
‫Right.

569
00:33:20,600 --> 00:33:26,270
‫Based on that, you can just change your database engine for every table based on the needs.

570
00:33:26,270 --> 00:33:30,110
‫If you have a table that only get queried, make it column, for example.

571
00:33:30,410 --> 00:33:30,680
‫Right.

572
00:33:30,800 --> 00:33:33,920
‫If you're doing a lot of analytics on it, make it a call.

573
00:33:34,100 --> 00:33:34,350
‫Right.

574
00:33:34,550 --> 00:33:38,660
‫If you're doing a lot of rights, probably you want to make it a raw store.

575
00:33:38,660 --> 00:33:39,050
‫Right.

576
00:33:39,470 --> 00:33:47,900
‫And think about all of this stuff like you cannot, for example, join a road based store, call a table

577
00:33:48,080 --> 00:33:50,510
‫with a column based table.

578
00:33:50,510 --> 00:33:50,960
‫Right.

579
00:33:51,600 --> 00:33:52,580
‫That's just bad.

580
00:33:52,850 --> 00:33:58,100
‫I mean, some day there is a support, but give them give them a break, guys.

581
00:33:58,100 --> 00:34:01,220
‫Give them a break, because that will be really disastrous.

582
00:34:01,220 --> 00:34:03,220
‫I can't even imagine what the database will do.

583
00:34:03,230 --> 00:34:05,690
‫It was going to freak out, I guess our guys.

584
00:34:05,980 --> 00:34:06,860
‫That's it for me today.

585
00:34:07,010 --> 00:34:09,110
‫What do you think about these two puppies?

586
00:34:09,320 --> 00:34:11,510
‫Let me know in the comments section below.

587
00:34:11,540 --> 00:34:13,820
‫I'm going to see you in the next one.

588
00:34:14,090 --> 00:34:14,840
‫You guys, they awesome.