PyConZA 2012: "Fractal Architectures" by Laurens Van Houtven

Slide 1

Slide 1 text

Fractal Architectures @lvh Friday, October 12, 2012

Slide 2

Slide 2 text

$ whoami Friday, October 12, 2012

Slide 3

Slide 3 text

@lvh, [email protected] Friday, October 12, 2012

Slide 4

Slide 4 text

Friday, October 12, 2012

Slide 5

Slide 5 text

#python Friday, October 12, 2012

Slide 6

Slide 6 text

Friday, October 12, 2012

Slide 7

Slide 7 text

Friday, October 12, 2012

Slide 8

Slide 8 text

Architectures Friday, October 12, 2012

Slide 9

Slide 9 text

Load Balancing Application Servers Cache Database Services Friday, October 12, 2012

Slide 10

Slide 10 text

LB App DB App App App DB Friday, October 12, 2012

Slide 11

Slide 11 text

! Friday, October 12, 2012

Slide 12

Slide 12 text

DB Friday, October 12, 2012

Slide 13

Slide 13 text

DB Friday, October 12, 2012

Slide 14

Slide 14 text

DB Friday, October 12, 2012

Slide 15

Slide 15 text

DB Friday, October 12, 2012

Slide 16

Slide 16 text

DB $ $$ Friday, October 12, 2012

Slide 17

Slide 17 text

App Friday, October 12, 2012

Slide 18

Slide 18 text

+ Queue Workers Friday, October 12, 2012

Slide 19

Slide 19 text

LB App DB App App App DB Friday, October 12, 2012

Slide 20

Slide 20 text

Friday, October 12, 2012

Slide 21

Slide 21 text

Friday, October 12, 2012

Slide 22

Slide 22 text

Friday, October 12, 2012

Slide 23

Slide 23 text

Friday, October 12, 2012

Slide 24

Slide 24 text

Friday, October 12, 2012

Slide 25

Slide 25 text

Friday, October 12, 2012

Slide 26

Slide 26 text

Friday, October 12, 2012

Slide 27

Slide 27 text

Friday, October 12, 2012

Slide 28

Slide 28 text

app Friday, October 12, 2012

Slide 29

Slide 29 text

user Friday, October 12, 2012

Slide 30

Slide 30 text

thing Friday, October 12, 2012

Slide 31

Slide 31 text

k n⋅k Friday, October 12, 2012

Slide 32

Slide 32 text

Friday, October 12, 2012

Slide 33

Slide 33 text

Friday, October 12, 2012

Slide 34

Slide 34 text

Friday, October 12, 2012

Slide 35

Slide 35 text

Friday, October 12, 2012

Slide 36

Slide 36 text

Restrictions Creativity Friday, October 12, 2012

Slide 37

Slide 37 text

Restrictions Strength Friday, October 12, 2012

Slide 38

Slide 38 text

Friday, October 12, 2012

Slide 39

Slide 39 text

! Friday, October 12, 2012

Slide 40

Slide 40 text

! Friday, October 12, 2012

Slide 41

Slide 41 text

Friday, October 12, 2012

Slide 42

Slide 42 text

Friday, October 12, 2012

Slide 43

Slide 43 text

Friday, October 12, 2012

Slide 44

Slide 44 text

Friday, October 12, 2012

Slide 45

Slide 45 text

Friday, October 12, 2012

Slide 46

Slide 46 text

Friday, October 12, 2012

Slide 47

Slide 47 text

Friday, October 12, 2012

Slide 48

Slide 48 text

Friday, October 12, 2012

Slide 49

Slide 49 text

Friday, October 12, 2012

Slide 50

Slide 50 text

Friday, October 12, 2012

Slide 51

Slide 51 text

Friday, October 12, 2012

Slide 52

Slide 52 text

Friday, October 12, 2012

Slide 53

Slide 53 text

SMTP Friday, October 12, 2012

Slide 54

Slide 54 text

import this Friday, October 12, 2012

Slide 55

Slide 55 text

Special cases aren't special enough to break the rules. Friday, October 12, 2012

Slide 56

Slide 56 text

Although practicality beats purity. Friday, October 12, 2012

Slide 57

Slide 57 text

Friday, October 12, 2012

Slide 58

Slide 58 text

Friday, October 12, 2012

Slide 59

Slide 59 text

Friday, October 12, 2012

Slide 60

Slide 60 text

Friday, October 12, 2012

Slide 61

Slide 61 text

Friday, October 12, 2012

Slide 62

Slide 62 text

Friday, October 12, 2012

Slide 63

Slide 63 text

Axiom Friday, October 12, 2012

Slide 64

Slide 64 text

Simple Friday, October 12, 2012

Slide 65

Slide 65 text

T O Friday, October 12, 2012

Slide 66

Slide 66 text

Friday, October 12, 2012

Slide 67

Slide 67 text

in-memory Friday, October 12, 2012

Slide 68

Slide 68 text

Sophisticated Friday, October 12, 2012

Slide 69

Slide 69 text

v1 (schema) v2 (schema) Friday, October 12, 2012

Slide 70

Slide 70 text

v1 x (reference) v2 x (reference) y (integer) Friday, October 12, 2012

Slide 71

Slide 71 text

v1 x (reference) y (text) v2 x (reference) y (text) Friday, October 12, 2012

Slide 72

Slide 72 text

A, vx ... A, vx+1 ... B, v1 ... Friday, October 12, 2012

Slide 73

Slide 73 text

A, vx ... B, vy ... C, vz ... Friday, October 12, 2012

Slide 74

Slide 74 text

ConstraintError: attribute [I.t = text()] must be (unicode string without NULL bytes); not 'str' Friday, October 12, 2012

Slide 75

Slide 75 text

Friday, October 12, 2012

Slide 76

Slide 76 text

+ Friday, October 12, 2012

Slide 77

Slide 77 text

Friday, October 12, 2012

Slide 78

Slide 78 text

Friday, October 12, 2012

Slide 79

Slide 79 text

Friday, October 12, 2012

Slide 80

Slide 80 text

! Friday, October 12, 2012

Slide 81

Slide 81 text

Alternatives Friday, October 12, 2012

Slide 82

Slide 82 text

Friday, October 12, 2012

Slide 83

Slide 83 text

Friday, October 12, 2012

Slide 84

Slide 84 text

Friday, October 12, 2012

Slide 85

Slide 85 text

Longer term storage Friday, October 12, 2012

Slide 86

Slide 86 text

Why? Friday, October 12, 2012

Slide 87

Slide 87 text

Friday, October 12, 2012

Slide 88

Slide 88 text

Friday, October 12, 2012

Slide 89

Slide 89 text

Server A Server B Long Term Storage Friday, October 12, 2012

Slide 90

Slide 90 text

Server A Server B Long Term Storage Friday, October 12, 2012

Slide 91

Slide 91 text

Server A Server B Long Term Storage Friday, October 12, 2012

Slide 92

Slide 92 text

Options? Friday, October 12, 2012

Slide 93

Slide 93 text

Friday, October 12, 2012

Slide 94

Slide 94 text

Friday, October 12, 2012

Slide 95

Slide 95 text

Friday, October 12, 2012

Slide 96

Slide 96 text

Friday, October 12, 2012

Slide 97

Slide 97 text

manhole Friday, October 12, 2012

Slide 98

Slide 98 text

Anything will work Friday, October 12, 2012

Slide 99

Slide 99 text

Friday, October 12, 2012

Slide 100

Slide 100 text

Friday, October 12, 2012

Slide 101

Slide 101 text

Friday, October 12, 2012

Slide 102

Slide 102 text

Friday, October 12, 2012

Slide 103

Slide 103 text

Friday, October 12, 2012

Slide 104

Slide 104 text

Friday, October 12, 2012

Slide 105

Slide 105 text

Friday, October 12, 2012

Slide 106

Slide 106 text

Friday, October 12, 2012

Slide 107

Slide 107 text

Two ways to better performance Friday, October 12, 2012

Slide 108

Slide 108 text

Do less work or Do it faster Friday, October 12, 2012

Slide 109

Slide 109 text

1E+00 1E+01 1E+02 1E+03 1E+04 1E+05 1E+06 1E+07 1E+08 1E+09 L1 L2 RAM SSD LAN HD CA-NL Friday, October 12, 2012

Slide 110

Slide 110 text

Local Low latency Friday, October 12, 2012

Slide 111

Slide 111 text

Query latency vs. Access latency Friday, October 12, 2012

Slide 112

Slide 112 text

(Almost) never Friday, October 12, 2012

Slide 113

Slide 113 text

vs Remote Local Friday, October 12, 2012

Slide 114

Slide 114 text

vs Remote Friday, October 12, 2012

Slide 115

Slide 115 text

. . . Friday, October 12, 2012

Slide 116

Slide 116 text

. . . { # Friday, October 12, 2012

Slide 117

Slide 117 text

Generalizations are always wrong, but... Friday, October 12, 2012

Slide 118

Slide 118 text

Computation is only rarely the bottleneck Friday, October 12, 2012

Slide 119

Slide 119 text

Data in the right place at the right time is Friday, October 12, 2012

Slide 120

Slide 120 text

pip install axiombench Friday, October 12, 2012

Slide 121

Slide 121 text

Friday, October 12, 2012

Slide 122

Slide 122 text

Some of my data doesn’t ﬁt in :-( Friday, October 12, 2012

Slide 123

Slide 123 text

Friday, October 12, 2012

Slide 124

Slide 124 text

Friday, October 12, 2012

Slide 125

Slide 125 text

Friday, October 12, 2012

Slide 126

Slide 126 text

Friday, October 12, 2012

Slide 127

Slide 127 text

Friday, October 12, 2012

Slide 128

Slide 128 text

Friday, October 12, 2012

Slide 129

Slide 129 text

sum(xs) / len(xs) Friday, October 12, 2012

Slide 130

Slide 130 text

users.aggregate, not user.aggregate Friday, October 12, 2012

Slide 131

Slide 131 text

Friday, October 12, 2012

Slide 132

Slide 132 text

Friday, October 12, 2012

Slide 133

Slide 133 text

world.data, not user.data Friday, October 12, 2012

Slide 134

Slide 134 text

Friday, October 12, 2012

Slide 135

Slide 135 text

Friday, October 12, 2012

Slide 136

Slide 136 text

duplication delegation separation Friday, October 12, 2012

Slide 137

Slide 137 text

Friday, October 12, 2012

Slide 138

Slide 138 text

Friday, October 12, 2012

Slide 139

Slide 139 text

Everything local :-) Friday, October 12, 2012

Slide 140

Slide 140 text

Simple :-) Friday, October 12, 2012

Slide 141

Slide 141 text

Duplication :-( Friday, October 12, 2012

Slide 142

Slide 142 text

Size? Friday, October 12, 2012

Slide 143

Slide 143 text

Updates? Friday, October 12, 2012

Slide 144

Slide 144 text

Limited data Limited mutation Friday, October 12, 2012

Slide 145

Slide 145 text

Friday, October 12, 2012

Slide 146

Slide 146 text

Friday, October 12, 2012

Slide 147

Slide 147 text

No duplication :-) Friday, October 12, 2012

Slide 148

Slide 148 text

Complex :-( Friday, October 12, 2012

Slide 149

Slide 149 text

Sometimes local Friday, October 12, 2012

Slide 150

Slide 150 text

Usage highly correlated with particular user Friday, October 12, 2012

Slide 151

Slide 151 text

Example: sales data, sharded by area Friday, October 12, 2012

Slide 152

Slide 152 text

Most queries local Friday, October 12, 2012

Slide 153

Slide 153 text

Aggregate queries cross shards predictably Friday, October 12, 2012

Slide 154

Slide 154 text

Tight coupling :-( Friday, October 12, 2012

Slide 155

Slide 155 text

Friday, October 12, 2012

Slide 156

Slide 156 text

Friday, October 12, 2012

Slide 157

Slide 157 text

Friday, October 12, 2012

Slide 158

Slide 158 text

Friday, October 12, 2012

Slide 159

Slide 159 text

Friday, October 12, 2012

Slide 160

Slide 160 text

Querying over all data is hard :-( Friday, October 12, 2012

Slide 161

Slide 161 text

Querying across stores is hard :-( Friday, October 12, 2012

Slide 162

Slide 162 text

Yes. Friday, October 12, 2012

Slide 163

Slide 163 text

Data analytics & Transactions Friday, October 12, 2012

Slide 164

Slide 164 text

Data analytics Friday, October 12, 2012

Slide 165

Slide 165 text

Most datasets Separation Friday, October 12, 2012

Slide 166

Slide 166 text

Friday, October 12, 2012

Slide 167

Slide 167 text

Big Data Specialized tools Friday, October 12, 2012

Slide 168

Slide 168 text

Friday, October 12, 2012

Slide 169

Slide 169 text

Friday, October 12, 2012

Slide 170

Slide 170 text

Missing sweet spot? Friday, October 12, 2012

Slide 171

Slide 171 text

Maybe. Friday, October 12, 2012

Slide 172

Slide 172 text

Transactions Friday, October 12, 2012

Slide 173

Slide 173 text

Cross-database transactions do exist Friday, October 12, 2012

Slide 174

Slide 174 text

ATTACH DATABASE Friday, October 12, 2012

Slide 175

Slide 175 text

dblink Friday, October 12, 2012

Slide 176

Slide 176 text

Do it in Python! Friday, October 12, 2012

Slide 177

Slide 177 text

Paxos Friday, October 12, 2012

Slide 178

Slide 178 text

Friday, October 12, 2012

Slide 179

Slide 179 text

Friday, October 12, 2012

Slide 180

Slide 180 text

No existing tools or frameworks :-( Friday, October 12, 2012

Slide 181

Slide 181 text

Friday, October 12, 2012

Slide 182

Slide 182 text

Friday, October 12, 2012

Slide 183

Slide 183 text

Friday, October 12, 2012

Slide 184

Slide 184 text

Friday, October 12, 2012

Slide 185

Slide 185 text

Be the change you want to see Friday, October 12, 2012

Slide 186

Slide 186 text

Divmod Mantissa Friday, October 12, 2012

Slide 187

Slide 187 text

exponent, maxims Friday, October 12, 2012

Slide 188

Slide 188 text

Friday, October 12, 2012