Slide 1

Slide 1 text

Scaling Science [email protected] Dr. Matt Wood

Slide 2

Slide 2 text

Hello

Slide 3

Slide 3 text

Science

Slide 4

Slide 4 text

Beautiful, unique.

Slide 5

Slide 5 text

Impossible to re-create

Slide 6

Slide 6 text

Snowflake Science

Slide 7

Slide 7 text

Reproducibility

Slide 8

Slide 8 text

Reproducibility scales science

Slide 9

Slide 9 text

Reproduce. Reuse. Remix.

Slide 10

Slide 10 text

Value++

Slide 11

Slide 11 text

No content

Slide 12

Slide 12 text

How do we get from here to there? 5PRINCIPLES REPRODUCIBILITY OF

Slide 13

Slide 13 text

1. Data has Gravity 5 PRINCIPLES REPRODUCIBILITY OF

Slide 14

Slide 14 text

Increasingly large data collections

Slide 15

Slide 15 text

1000 Genomes Project: 200Tb

Slide 16

Slide 16 text

Challenging to obtain and manage

Slide 17

Slide 17 text

Expensive to experiment

Slide 18

Slide 18 text

Large barrier to reproducibility

Slide 19

Slide 19 text

Data size will increase

Slide 20

Slide 20 text

Data integration will increase

Slide 21

Slide 21 text

Data dependencies will increase

Slide 22

Slide 22 text

Move data to the users

Slide 23

Slide 23 text

Move data to the users X

Slide 24

Slide 24 text

Move tools to the data

Slide 25

Slide 25 text

Place data where it can consumed by tools

Slide 26

Slide 26 text

Place tools where they can access data

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

No content

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

Canonical source

Slide 31

Slide 31 text

No content

Slide 32

Slide 32 text

More data, more users, more uses, more locations

Slide 33

Slide 33 text

Cost

Slide 34

Slide 34 text

Force multiplier

Slide 35

Slide 35 text

Cost

Slide 36

Slide 36 text

Complexity

Slide 37

Slide 37 text

Cost and complexity kill reproducibility

Slide 38

Slide 38 text

Utility computing

Slide 39

Slide 39 text

Availability

Slide 40

Slide 40 text

Pay-as-you-go

Slide 41

Slide 41 text

Flexibility

Slide 42

Slide 42 text

Performance

Slide 43

Slide 43 text

CPU + IO

Slide 44

Slide 44 text

Intel Xeon E5 NVIDIA Tesla GPUs

Slide 45

Slide 45 text

240 TFLOPS

Slide 46

Slide 46 text

90 - 120k IOPS on SSDs

Slide 47

Slide 47 text

Performance through productivity

Slide 48

Slide 48 text

Cost

Slide 49

Slide 49 text

On-demand access

Slide 50

Slide 50 text

Reserved capacity

Slide 51

Slide 51 text

100% Reserved capacity

Slide 52

Slide 52 text

100% Reserved capacity On-demand

Slide 53

Slide 53 text

100% Reserved capacity On-demand

Slide 54

Slide 54 text

Spot instances

Slide 55

Slide 55 text

Utility computing enhanced reproducibility

Slide 56

Slide 56 text

No content

Slide 57

Slide 57 text

2. Ease of use is a pre-requisite 5 PRINCIPLES REPRODUCIBILITY OF

Slide 58

Slide 58 text

http://headrush.typepad.com/creating_passionate_users/2005/10/getting_users_p.html

Slide 59

Slide 59 text

Help overcome the suck threshold

Slide 60

Slide 60 text

Easy to embrace and extend

Slide 61

Slide 61 text

Choose the right abstraction for the user

Slide 62

Slide 62 text

$ ec2-run-instances

Slide 63

Slide 63 text

$ starcluster start

Slide 64

Slide 64 text

No content

Slide 65

Slide 65 text

Package and automate

Slide 66

Slide 66 text

Package and automate Amazon machine images, VM import

Slide 67

Slide 67 text

Package and automate Amazon machine images, VM import Deployment scripts, CloudFormation, Chef, Puppet

Slide 68

Slide 68 text

Expert-as-a-service

Slide 69

Slide 69 text

No content

Slide 70

Slide 70 text

No content

Slide 71

Slide 71 text

1000 Genomes Cloud BioLinux

Slide 72

Slide 72 text

No content

Slide 73

Slide 73 text

Your HiSeq data Illumina BaseSpace

Slide 74

Slide 74 text

Architectural freedom

Slide 75

Slide 75 text

Freedom of abstraction

Slide 76

Slide 76 text

3. Reuse is as important as reproduction 5 PRINCIPLES REPRODUCIBILITY OF

Slide 77

Slide 77 text

Seven Deadly sins of Bioinformatics: http://www.slideshare.net/dullhunk/the-seven-deadly-sins-of-bioinformatics

Slide 78

Slide 78 text

Seven Deadly sins of Bioinformatics: http://www.slideshare.net/dullhunk/the-seven-deadly-sins-of-bioinformatics

Slide 79

Slide 79 text

Infonauts are hackers

Slide 80

Slide 80 text

They have their own way of working

Slide 81

Slide 81 text

The ‘Big Red Button’

Slide 82

Slide 82 text

Fire and forget reproduction is a good first step, but limits longer term value.

Slide 83

Slide 83 text

Monolithic, one-stop-shop

Slide 84

Slide 84 text

Work well for intended purpose

Slide 85

Slide 85 text

Challenging to install, dependency heavy

Slide 86

Slide 86 text

Di cult to grok

Slide 87

Slide 87 text

Inflexible

Slide 88

Slide 88 text

Infonauts are hackers: embrace it.

Slide 89

Slide 89 text

Small things. Loosely coupled.

Slide 90

Slide 90 text

Easier to grok

Slide 91

Slide 91 text

Easier to reuse

Slide 92

Slide 92 text

Easier to integrate

Slide 93

Slide 93 text

Lower barrier to entry

Slide 94

Slide 94 text

Scale out

Slide 95

Slide 95 text

Build for reuse. Be remix friendly. Maximize value.

Slide 96

Slide 96 text

4. Build for collaboration 5 PRINCIPLES REPRODUCIBILITY OF

Slide 97

Slide 97 text

Workflows are memes

Slide 98

Slide 98 text

Reproduction is just the first step

Slide 99

Slide 99 text

Bill of materials: code, data, configuration, infrastructure

Slide 100

Slide 100 text

Full definition for reproduction

Slide 101

Slide 101 text

Utility computing provides a playground for bioinformatics

Slide 102

Slide 102 text

Code + AMI + custom datasets + public datasets + databases + compute + result data

Slide 103

Slide 103 text

Code + AMI + custom datasets + public datasets + databases + compute + result data

Slide 104

Slide 104 text

Code + AMI + custom datasets + public datasets + databases + compute + result data

Slide 105

Slide 105 text

Code + AMI + custom datasets + public datasets + databases + compute + result data

Slide 106

Slide 106 text

Package, automate, contribute.

Slide 107

Slide 107 text

Utility platform provides scale for production runs

Slide 108

Slide 108 text

Drug discovery on 50k cores: Less than $1000

Slide 109

Slide 109 text

5. Provenance is a first class object 5 PRINCIPLES REPRODUCIBILITY OF

Slide 110

Slide 110 text

Versioning becomes really important

Slide 111

Slide 111 text

Especially in an active community

Slide 112

Slide 112 text

Doubly so with loosely coupled tools

Slide 113

Slide 113 text

Provenance metadata is a first class entity

Slide 114

Slide 114 text

Distributed provenance

Slide 115

Slide 115 text

1. Data has gravity 2. Ease of use is a pre-requisite 3. Reuse is as important as reproduction 4. Build for collaboration 5. Provenance is a first class object 5PRINCIPLES REPRODUCIBILITY OF

Slide 116

Slide 116 text

No content

Slide 117

Slide 117 text

Thank you aws.amazon.com @mza [email protected]