Efficiency and Rock 'n’ Roll (Really!)

by Holly Cummins

Slide 1

Slide 1 text

Holly Cummins Red Hat JNation 2025 Efficiency & Rock ’n’ Roll

Slide 2

Slide 2 text

@hollycummins.com #Quarkus late nights

Slide 3

Slide 3 text

@hollycummins.com #Quarkus late nights

Slide 4

Slide 4 text

@hollycummins.com #Quarkus stimulants

Slide 5

Slide 5 text

@hollycummins.com #Quarkus *@!#f&*%&!

Slide 6

Slide 6 text

@hollycummins.com #Quarkus // wtf

Slide 7

Slide 7 text

@hollycummins.com #Quarkus programming ≈ rock ’n' roll

Slide 8

Slide 8 text

@hollycummins.com #Quarkus programming ≈ rock ’n' roll efficiency rock ’n' roll

Slide 9

Slide 9 text

@hollycummins.com #Quarkus programming ≈ rock ’n' roll efficiency rock ’n' roll?? …

Slide 10

Slide 10 text

@hollycummins.com #Quarkus efficiency achieving the largest amount of useful work using as little energy and effort as possible

Slide 11

Slide 11 text

@hollycummins.com #Quarkus

Slide 12

Slide 12 text

@hollycummins.com #Quarkus rock & roll a style of popular music played on electronically amplified instruments and characterized by a persistent heavily accented beat, repetition of simple phrases …

Slide 13

Slide 13 text

@hollycummins.com #Quarkus rock & roll …

Slide 14

Slide 14 text

@hollycummins.com #Quarkus coolness

Slide 15

Slide 15 text

@hollycummins.com #Quarkus turning it up to 11

Slide 16

Slide 16 text

@hollycummins.com #Quarkus productive

Slide 17

Slide 17 text

@hollycummins.com #Quarkus useful

Slide 18

Slide 18 text

@hollycummins.com #Quarkus excess

Slide 19

Slide 19 text

@hollycummins.com #Quarkus

Slide 20

Slide 20 text

@hollycummins.com #Quarkus minimising waste

Slide 21

Slide 21 text

@hollycummins.com #Quarkus minimising waste

Slide 22

Slide 22 text

@hollycummins.com #Quarkus

Slide 23

Slide 23 text

@hollycummins.com #Quarkus rage against the machine

Slide 24

Slide 24 text

@hollycummins.com #Quarkus rage against the machine

Slide 25

Slide 25 text

@hollycummins.com #Quarkus uh oh.

Slide 26

Slide 26 text

senior principal software engineer quarkus

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

No content

Slide 29

Slide 29 text

@hollycummins.com #Quarkus

Slide 30

Slide 30 text

@hollycummins.com #Quarkus queueing theory

Slide 31

Slide 31 text

@hollycummins.com #Quarkus queueing theory

Slide 32

Slide 32 text

@hollycummins.com #Quarkus being lazy

Slide 33

Slide 33 text

@hollycummins.com #Quarkus being a rebel

Slide 34

Slide 34 text

@hollycummins.com 1700s machine

Slide 35

Slide 35 text

@hollycummins.com 1700s machine energy

Slide 36

Slide 36 text

@hollycummins.com 1700s machine energy energy

Slide 37

Slide 37 text

@hollycummins.com 1700s machine energy energy useful

Slide 38

Slide 38 text

@hollycummins.com 1700s waste machine energy energy useful

Slide 39

Slide 39 text

@hollycummins.com 1890s processes

Slide 40

Slide 40 text

@hollycummins.com 1890s processes

Slide 41

Slide 41 text

@hollycummins.com time money 1890s processes

Slide 42

Slide 42 text

@hollycummins.com time money 1890s value processes

Slide 43

Slide 43 text

@hollycummins.com time money 1890s value processes waste

Slide 44

Slide 44 text

@hollycummins.com software 1960s

Slide 45

Slide 45 text

@hollycummins.com software time electricity hardware 1960s

Slide 46

Slide 46 text

@hollycummins.com software time electricity hardware answers 1960s

Slide 47

Slide 47 text

@hollycummins.com software time electricity hardware answers 1960s waste

Slide 48

Slide 48 text

@hollycummins.com #Quarkus waste

Slide 49

Slide 49 text

@hollycummins.com #Quarkus 300 years of practice

Slide 50

Slide 50 text

@hollycummins.com #Quarkus “this provisioning software is broken”

Slide 51

Slide 51 text

@hollycummins.com #Quarkus what we sold “this provisioning software is broken” 10 minute provision-time

Slide 52

Slide 52 text

@hollycummins.com #Quarkus what we sold “this provisioning software is broken” 10 minute provision-time 3 month provision- time what the client thought they’d got

Slide 53

Slide 53 text

@hollycummins.com #Quarkus what we sold “this provisioning software is broken” 10 minute provision-time 3 month provision- time what the client thought they’d got the reason 84-step pre-approval process

Slide 54

Slide 54 text

@hollycummins.com

Slide 55

Slide 55 text

@hollycummins.com

Slide 56

Slide 56 text

@hollycummins.com

Slide 57

Slide 57 text

@hollycummins.com

Slide 58

Slide 58 text

@hollycummins.com

Slide 59

Slide 59 text

@hollycummins.com

Slide 60

Slide 60 text

@hollycummins.com doom zoom

Slide 61

Slide 61 text

@hollycummins.com doom zoom landing page

Slide 62

Slide 62 text

@hollycummins.com doom zoom landing page

Slide 63

Slide 63 text

@hollycummins.com doom zoom landing page

Slide 64

Slide 64 text

@hollycummins.com https://blog.linkedin.com/2017/august/3/making-linkedin-more-accessible-via-linkedin-lite Nashik

Slide 65

Slide 65 text

@hollycummins.com https://blog.linkedin.com/2017/august/3/making-linkedin-more-accessible-via-linkedin-lite “my heart sank … our new feature failed to load because of poor internet connectivity” Nashik

Slide 66

Slide 66 text

@hollycummins.com https://blog.linkedin.com/2017/august/3/making-linkedin-more-accessible-via-linkedin-lite modern web is so inefficient it is useless for part of its audience “my heart sank … our new feature failed to load because of poor internet connectivity” Nashik

Slide 67

Slide 67 text

@hollycummins.com code waste is killing the planet

Slide 68

Slide 68 text

@hollycummins.com forgotten code is killing the planet

Slide 69

Slide 69 text

25% of 16,000 servers doing no useful work https://www.anthesisgroup.com/wp-content/uploads/2019/11/Comatose-Servers-Redux-2017.pdf zombie servers

Slide 70

Slide 70 text

the average server: 12 - 18% of capacity https://www.nrdc.org/sites/default/files/data-center-efficiency-assessment-IB.pdf

Slide 71

Slide 71 text

@hollycummins.com unused data is killing the planet

Slide 72

Slide 72 text

68% of data is stored and never used https://www.seagate.com/gb/en/news/news-archive/seagates-rethink-data-report-reveals-that-68-percent-of-data-available-to-businesses-goes-unleveraged-pr-master/

Slide 73

Slide 73 text

@hollycummins.com it’s not just money

Slide 74

Slide 74 text

@hollycummins.com it’s not just electricity it’s not just money

Slide 75

Slide 75 text

@hollycummins.com it’s not just electricity it’s not just money it’s embodied carbon

Slide 76

Slide 76 text

@hollycummins.com it’s not just electricity it’s water it’s not just money it’s embodied carbon

Slide 77

Slide 77 text

@hollycummins.com it’s not just electricity it’s water it’s e-waste it’s not just money it’s embodied carbon

Slide 78

Slide 78 text

@hollycummins.com we have solutions good news, everyone

Slide 79

Slide 79 text

@hollycummins.com solution: LightSwitchOps • making turning servers off as safe and easy as turning lights off. • and then automate it.

Slide 80

Slide 80 text

absurdly simple scripts my shell script to power down

Slide 81

Slide 81 text

we used to leave our applications running all the time @darkandnerdy simple automation

Slide 82

Slide 82 text

we used to leave our applications running all the time when we scripted turning them off at night, we reduced our cloud bill by 30% @darkandnerdy simple automation

Slide 83

Slide 83 text

@hollycummins.com #Quarkus fancier scripts … with a frontend and a backend

Slide 84

Slide 84 text

@hollycummins.com #Quarkus why does low utilisation happen?

Slide 85

Slide 85 text

@hollycummins.com #Quarkus why does low utilisation happen? forgetfulness

Slide 86

Slide 86 text

@hollycummins.com #Quarkus apathy why does low utilisation happen? forgetfulness

Slide 87

Slide 87 text

@hollycummins.com #Quarkus apathy why does low utilisation happen? forgetfulness fear

Slide 88

Slide 88 text

@hollycummins.com #Quarkus be a rebel

Slide 89

Slide 89 text

@hollycummins.com #Quarkus be a rebel

Slide 90

Slide 90 text

@hollycummins.com #Quarkus saves energy saves money bonus: tests disaster recovery

Slide 91

Slide 91 text

@hollycummins.com #Quarkus saves energy saves money bonus: tests disaster recovery

Slide 92

Slide 92 text

@hollycummins.com #Quarkus solution: faster code

Slide 93

Slide 93 text

@hollycummins.com the vrroooom model

Slide 94

Slide 94 text

@hollycummins.com

Slide 95

Slide 95 text

@hollycummins.com invented by Dr. Vroom (really!)

Slide 96

Slide 96 text

@hollycummins.com naming is the hardest problem in computer science invented by Dr. Vroom (really!)

Slide 97

Slide 97 text

@hollycummins.com naming is the hardest problem in computer science

Slide 98

Slide 98 text

the wrong kind of rock ’n’ roll

Slide 99

Slide 99 text

the wrong kind of rock ’n’ roll

Slide 100

Slide 100 text

my vrroooom model

Slide 101

Slide 101 text

No content

Slide 102

Slide 102 text

@hollycummins.com #Quarkus

Slide 103

Slide 103 text

@hollycummins.com #Quarkus these two columns are almost the same

Slide 104

Slide 104 text

@hollycummins.com #Quarkus energy efficiency across programming languages Python Rust Java Go Javascript

Slide 105

Slide 105 text

@hollycummins.com #Quarkus the trend line is more or less straight energy efficiency across programming languages Python Rust Java Go Javascript

Slide 106

Slide 106 text

No content

Slide 107

Slide 107 text

case study quarkus

Slide 108

Slide 108 text

@hollycummins.com #Quarkus the distant past Java application frameworks needed dynamic dependencies

Slide 109

Slide 109 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 110

Slide 110 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 111

Slide 111 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 112

Slide 112 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 113

Slide 113 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 114

Slide 114 text

@hollycummins.com #Quarkus cloud apps are immutable now

Slide 115

Slide 115 text

@hollycummins.com #Quarkus a highly dynamic runtime in a container is pointless

Slide 116

Slide 116 text

@hollycummins.com #Quarkus java dynamism

Slide 117

Slide 117 text

@hollycummins.com #Quarkus java dynamism build time

Slide 118

Slide 118 text

@hollycummins.com #Quarkus java dynamism build time runtime

Slide 119

Slide 119 text

@hollycummins.com #Quarkus java dynamism build time runtime

Slide 120

Slide 120 text

@hollycummins.com #Quarkus java dynamism packaging (maven, gradle…) build time runtime

Slide 121

Slide 121 text

@hollycummins.com #Quarkus java dynamism build time runtime

Slide 122

Slide 122 text

@hollycummins.com #Quarkus java dynamism build time runtime

Slide 123

Slide 123 text

@hollycummins.com #Quarkus java dynamism build time runtime load and parse • config files • properties • yaml • xml • etc.

Slide 124

Slide 124 text

@hollycummins.com #Quarkus java dynamism build time runtime

Slide 125

Slide 125 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime • classpath scanning and annotation discovery • attempt to load class to enable/ disable features

Slide 126

Slide 126 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime

Slide 127

Slide 127 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime build a metamodel of the world

Slide 128

Slide 128 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime

Slide 129

Slide 129 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime start • thread pools • I/O • etc.

Slide 130

Slide 130 text

@hollycummins.com #Quarkus java dynamism @ @ build time runtime ready to do work!

Slide 131

Slide 131 text

what if we start the application more than once? @ @

Slide 132

Slide 132 text

what if we start the application more than once? @ @ @ @

Slide 133

Slide 133 text

what if we start the application more than once? @ @ @ @ @ @

Slide 134

Slide 134 text

what if we start the application more than once? @ @ @ @ @ @ @ @

Slide 135

Slide 135 text

what if we start the application more than once? @ @ @ @ @ @ @ @ so much work gets redone every time

Slide 136

Slide 136 text

@hollycummins.com Hibernate speed example: JTA auto-wiring

Slide 137

Slide 137 text

@hollycummins.com Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”);

Slide 138

Slide 138 text

@hollycummins.com Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”);

Slide 139

Slide 139 text

@hollycummins.com Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”);

Slide 140

Slide 140 text

Slide 141

Slide 141 text

@hollycummins.com Hibernate speed example: JTA auto-wiring Class.forName(“LikelyJTAImplementation”); Class.forName(“APossibleJTAImplementation”); Class.forName(“AnotherJTAImplementation”); Class.forName(“NicheJTAImplementation”); …

Slide 142

Slide 142 text

Slide 143

Slide 143 text

Slide 144

Slide 144 text

Slide 145

Slide 145 text

@hollycummins.com the true cost of loaded classes isn’t just memory + start time

Slide 146

Slide 146 text

@hollycummins.com the true cost of loaded classes isn’t just memory + start time method dispatching:

Slide 147

Slide 147 text

@hollycummins.com interface the true cost of loaded classes isn’t just memory + start time method dispatching:

Slide 148

Slide 148 text

@hollycummins.com unused implementation the one we want interface unused implementation unused implementation the true cost of loaded classes isn’t just memory + start time method dispatching:

Slide 149

Slide 149 text

@hollycummins.com unused implementation the one we want interface unused implementation unused implementation the true cost of loaded classes isn’t just memory + start time method dispatching:

Slide 150

Slide 150 text

@hollycummins.com unused implementation the one we want interface megamorphic call slow dispatching unused implementation unused implementation the true cost of loaded classes isn’t just memory + start time method dispatching:

Slide 151

Slide 151 text

@hollycummins.com the true cost of loaded classes isn’t just memory + start time the one we want interface

Slide 152

Slide 152 text

@hollycummins.com the true cost of loaded classes isn’t just memory + start time the one we want monomorphic call fast dispatching interface

Slide 153

Slide 153 text

@hollycummins.com how do we fix all this?

Slide 154

Slide 154 text

@ @ build time runtime what if we initialize at build time?

Slide 155

Slide 155 text

@ @ build time runtime what if we initialize at build time?

Slide 156

Slide 156 text

@ @ build time runtime ready to do work! what if we initialize at build time?

Slide 157

Slide 157 text

@ @ build time runtime ready to do work! start • thread pools • I/O • etc. what if we initialize at build time?

Slide 158

Slide 158 text

@ @ repeated starts are now efficient

Slide 159

Slide 159 text

@ @ repeated starts are now efficient

Slide 160

Slide 160 text

@ @ repeated starts are now efficient

Slide 161

Slide 161 text

@ @ repeated starts are now efficient

Slide 162

Slide 162 text

@ @ repeated starts are now efficient

Slide 163

Slide 163 text

@ @ repeated starts are now efficient

Slide 164

Slide 164 text

@ @ repeated starts are now efficient less wasted work

Slide 165

Slide 165 text

@hollycummins.com Setup: • REST + CRUD • large heap • RAPL energy measurement Assumptions: • US energy mix climate impact of framework choice

Slide 166

Slide 166 text

@hollycummins.com Setup: • REST + CRUD • large heap • RAPL energy measurement Assumptions: • US energy mix climate impact of framework choice shorter line means lower max throughput

Slide 167

Slide 167 text

@hollycummins.com Setup: • REST + CRUD • large heap • RAPL energy measurement Assumptions: • US energy mix climate impact of framework choice shorter line means lower max throughput higher line means worse carbon footprint

Slide 168

Slide 168 text

@hollycummins.com Setup: • REST + CRUD • large heap • RAPL energy measurement Assumptions: • US energy mix climate impact of framework choice vrrrooom model in action: quarkus on JVM has the smallest footprint … because it has the highest throughput shorter line means lower max throughput higher line means worse carbon footprint

Slide 169

Slide 169 text

@hollycummins.com throughput startup time + footprint

Slide 170

Slide 170 text

@hollycummins.com we beat the trade-off. throughput startup time + footprint

Slide 171

Slide 171 text

@hollycummins.com we beat the trade-off. throughput startup time + footprint it’s a double-win.

Slide 172

Slide 172 text

No content

Slide 173

Slide 173 text

@hollycummins.com we beat another trade-off. human efficiency machine efficiency

Slide 174

Slide 174 text

@hollycummins.com we beat another trade-off. human efficiency machine efficiency

Slide 175

Slide 175 text

@hollycummins.com we beat another trade-off. human efficiency machine efficiency another double-win.

Slide 176

Slide 176 text

@hollycummins.com boilerplate code

Slide 177

Slide 177 text

@hollycummins.com live coding

Slide 178

Slide 178 text

@hollycummins.com live coding -less energy wasted running full build + launch cycles

Slide 179

Slide 179 text

@hollycummins.com live coding -less energy wasted running full build + launch cycles -less dev time wasted waiting for full build + launch cycles

Slide 180

Slide 180 text

@hollycummins.com targeted continuous testing

Slide 181

Slide 181 text

@hollycummins.com targeted continuous testing -less energy wasted running full test suites

Slide 182

Slide 182 text

@hollycummins.com targeted continuous testing -less energy wasted running full test suites -less dev time wasted waiting for full test runs

Slide 183

Slide 183 text

@hollycummins.com #Quarkus developer joy

Slide 184

Slide 184 text

@hollycummins.com #Quarkus is a double win developer joy

Slide 185

Slide 185 text

@hollycummins.com #Quarkus the double win isn’t just a java thing.

Slide 186

Slide 186 text

@hollycummins.com #Quarkus the ai-lephant in the room

Slide 187

Slide 187 text

@hollycummins.com #Quarkus does developer efficiency even matter?

Slide 188

Slide 188 text

@hollycummins.com

Slide 189

Slide 189 text

@hollycummins.com

Slide 190

Slide 190 text

@hollycummins.com

Slide 191

Slide 191 text

@hollycummins.com

Slide 192

Slide 192 text

@hollycummins.com

Slide 193

Slide 193 text

@hollycummins.com

Slide 194

Slide 194 text

@hollycummins.com 70% unnecessary code

Slide 195

Slide 195 text

@hollycummins.com

Slide 196

Slide 196 text

@hollycummins.com we spend far more time reading code than writing it

Slide 197

Slide 197 text

@hollycummins.com shifting the work to future-you

Slide 198

Slide 198 text

@hollycummins.com illusion of efficiency

Slide 199

Slide 199 text

@hollycummins.com #Quarkus illusions of efficiency are everywhere

Slide 200

Slide 200 text

@hollycummins.com

Slide 201

Slide 201 text

@hollycummins.com

Slide 202

Slide 202 text

@hollycummins.com

Slide 203

Slide 203 text

@hollycummins.com even for machines, there is a limit to efficiency

Slide 204

Slide 204 text

@hollycummins.com source: http://geosci.uchicago.edu/~moyer/GEOS24705/2016/Assignments/PS7.pdf

Slide 205

Slide 205 text

@hollycummins.com thermodynamics source: http://geosci.uchicago.edu/~moyer/GEOS24705/2016/Assignments/PS7.pdf

Slide 206

Slide 206 text

@hollycummins.com

Slide 207

Slide 207 text

@hollycummins.com

Slide 208

Slide 208 text

@hollycummins.com theoretical max efficiency of a combustion engine

Slide 209

Slide 209 text

@hollycummins.com theoretical max efficiency of a combustion engine 37%

Slide 210

Slide 210 text

@hollycummins.com typical efficiency of a combustion engine

Slide 211

Slide 211 text

@hollycummins.com ~20% typical efficiency of a combustion engine

Slide 212

Slide 212 text

@hollycummins.com it’s detuned so the engine does not wear out ~20% typical efficiency of a combustion engine

Slide 213

Slide 213 text

@hollycummins.com it’s detuned so the engine does not wear out ~20% typical efficiency of a combustion engine

Slide 214

Slide 214 text

@hollycummins.com

Slide 215

Slide 215 text

@hollycummins.com highly efficient (optimum number of legs)

Slide 216

Slide 216 text

@hollycummins.com highly efficient (optimum number of legs) less efficient (more legs than needed)

Slide 217

Slide 217 text

@hollycummins.com highly efficient (optimum number of legs) less efficient (more legs than needed) resilient

Slide 218

Slide 218 text

@hollycummins.com highly efficient (optimum number of legs) less efficient (more legs than needed) resilient

Slide 219

Slide 219 text

@hollycummins.com highly efficient (optimum number of legs) no resiliency less efficient (more legs than needed) resilient

Slide 220

Slide 220 text

@hollycummins.com highly efficient (optimum number of legs) no resiliency less efficient (more legs than needed) resilient resiliency lowers efficiency

Slide 221

Slide 221 text

@hollycummins.com

Slide 222

Slide 222 text

@hollycummins.com speed

Slide 223

Slide 223 text

@hollycummins.com speed vs resiliency

Slide 224

Slide 224 text

@hollycummins.com it’s the same for people

Slide 225

Slide 225 text

@hollycummins.com it’s the same for people all work and no play is … not efficient

Slide 226

Slide 226 text

@hollycummins.com

Slide 227

Slide 227 text

@hollycummins.com hi colleague, could you ple-

Slide 228

Slide 228 text

@hollycummins.com hi colleague, could you ple- argh! &!*$*%@!{*%^!^! busy! NO! I CANNOT!

Slide 229

Slide 229 text

@hollycummins.com #Quarkus the value of slack

Slide 230

Slide 230 text

@hollycummins.com #Quarkus the value of slack

Slide 231

Slide 231 text

@hollycummins.com #Quarkus the value of slack

Slide 232

Slide 232 text

@hollycummins.com #Quarkus the value of slack

Slide 233

Slide 233 text

@hollycummins.com #Quarkus the value of slack naming is still the hardest problem in computer science

Slide 234

Slide 234 text

@hollycummins.com #Quarkus

Slide 235

Slide 235 text

@holly_cummins #RedHat doing nothing

Slide 236

Slide 236 text

@holly_cummins #RedHat doing nothing efficiency

Slide 237

Slide 237 text

@holly_cummins #RedHat the default mode network idle minds can solve hard problems

Slide 238

Slide 238 text

#RedHat @holly_cummins

Slide 239

Slide 239 text

@hollycummins.com #Quarkus

Slide 240

Slide 240 text

@hollycummins.com #Quarkus

Slide 241

Slide 241 text

@hollycummins.com #Quarkus 14% took showers specifically for the purpose of coming up with ideas

Slide 242

Slide 242 text

@hollycummins.com psychology says we need idle time; maths agrees

Slide 243

Slide 243 text

@hollycummins.com queueing theory says systems need to run under-capacity to function psychology says we need idle time; maths agrees

Slide 244

Slide 244 text

@hollycummins.com queuing theory basics queue arrival process servers completed work

Slide 245