Slide 1

Slide 1 text

There is no magic There is only awesome D e e p a k S i n g h A platform for data science

Slide 2

Slide 2 text

bioinformatics image: Ethan Hein

Slide 3

Slide 3 text

3

Slide 4

Slide 4 text

collection

Slide 5

Slide 5 text

curation

Slide 6

Slide 6 text

analysis

Slide 7

Slide 7 text

what’s the big deal?

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

Source: http://www.nature.com/news/specials/bigdata/index.html

Slide 10

Slide 10 text

Image: Yael Fitzpatrick (AAAS)

Slide 11

Slide 11 text

Image: Yael Fitzpatrick (AAAS)

Slide 12

Slide 12 text

lots of data

Slide 13

Slide 13 text

lots of people

Slide 14

Slide 14 text

lots of places

Slide 15

Slide 15 text

to make data effective

Slide 16

Slide 16 text

versioning

Slide 17

Slide 17 text

provenance

Slide 18

Slide 18 text

filter

Slide 19

Slide 19 text

aggregate

Slide 20

Slide 20 text

extend

Slide 21

Slide 21 text

mashup

Slide 22

Slide 22 text

human interfaces

Slide 23

Slide 23 text

No content

Slide 24

Slide 24 text

hard problem

Slide 25

Slide 25 text

really hard problem

Slide 26

Slide 26 text

No content

Slide 27

Slide 27 text

change how we think about compute

Slide 28

Slide 28 text

change how we think about data

Slide 29

Slide 29 text

change how we think about science

Slide 30

Slide 30 text

information platforms

Slide 31

Slide 31 text

Image: Drew Conway

Slide 32

Slide 32 text

dataspaces Further reading: Jeff Hammerbacher, Information Platforms and the rise of the data scientist, Beautiful Data

Slide 33

Slide 33 text

the unreasonable effectiveness of data Halevy, et al. IEEE Intelligent Systems, 24, 8-12 (2009)

Slide 34

Slide 34 text

accept all data formats

Slide 35

Slide 35 text

evolve APIs

Slide 36

Slide 36 text

beyond the database and the data warehouse

Slide 37

Slide 37 text

data as a programmable resource

Slide 38

Slide 38 text

data as a royal garden

Slide 39

Slide 39 text

compute as a fungible commodity

Slide 40

Slide 40 text

which brings us to ...

Slide 41

Slide 41 text

amazon web services

Slide 42

Slide 42 text

No content

Slide 43

Slide 43 text

No content

Slide 44

Slide 44 text

common characteristics

Slide 45

Slide 45 text

on demand

Slide 46

Slide 46 text

pay as you go

Slide 47

Slide 47 text

secure

Slide 48

Slide 48 text

elastic

Slide 49

Slide 49 text

3000 CPU’s for one firm’s risk management application !"#$%&'()'*+,'-./01.2%/' 344'+567/'(.' 8%%9%.:/' ;<"&/:1=' >?,3?,44@' A&B:1=' >?,>?,44@' C".:1=' >?,D?,44@' E(.:1=' >?,F?,44@' ;"%/:1=' >?,G?,44@' C10"&:1=' >?,H?,44@' I%:.%/:1=' >?,,?,44@' 3444JJ' 344'JJ'

Slide 50

Slide 50 text

programmable

Slide 51

Slide 51 text

“infrastructure as code”

Slide 52

Slide 52 text

No content

Slide 53

Slide 53 text

include_recipe "packages" include_recipe "ruby" include_recipe "apache2" if platform?("centos","redhat") if dist_only? # just the gem, we'll install the apache module within apache2 package "rubygem-passenger" return else package "httpd-devel" end else %w{ apache2-prefork-dev libapr1-dev }.each do |pkg| package pkg do action :upgrade end end end gem_package "passenger" do version node[:passenger][:version] end execute "passenger_module" do command 'echo -en "\n\n\n\n" | passenger-install-apache2-module' creates node[:passenger][:module_path] end

Slide 54

Slide 54 text

import boto import boto.emr from boto.emr.step import StreamingStep from boto.emr.bootstrap_action import BootstrapAction import time # set your aws keys and S3 bucket, e.g. from environment or .boto AWSKEY= SECRETKEY= S3_BUCKET= NUM_INSTANCES = 1 conn = boto.connect_emr(AWSKEY,SECRETKEY) bootstrap_step = BootstrapAction("download.tst", "s3://elasticmapreduce/bootstrap-actions/download.sh",None) step = StreamingStep(name='Wordcount', mapper='s3n://elasticmapreduce/samples/wordcount/wordSplitter.py', cache_files = ["s3n://" + S3_BUCKET + "/boto.mod#boto.mod"], reducer='aggregate', input='s3n://elasticmapreduce/samples/wordcount/input', output='s3n://' + S3_BUCKET + '/output/wordcount_output') jobid = conn.run_jobflow( name="testbootstrap", log_uri="s3://" + S3_BUCKET + "/logs", steps = [step], bootstrap_actions=[bootstrap_step], num_instances=NUM_INSTANCES) print "finished spawning job (note: starting still takes time)" state = conn.describe_jobflow(jobid).state print "job state = ", state print "job id = ", jobid while state != u'COMPLETED': print time.localtime() time.sleep(30) state = conn.describe_jobflow(jobid).state print "job state = ", state print "job id = ", jobid print "final output can be found in s3://" + S3_BUCKET + "/output" + TIMESTAMP print "try: $ s3cmd sync s3://" + S3_BUCKET + "/output" + TIMESTAMP + " ." Connect to Elastic MapReduce Install packages Set up mappers & reduces job state

Slide 55

Slide 55 text

No content

Slide 56

Slide 56 text

“I terminate the instance and relaunch it. Thats my error handling” Source: @jtimberman on Twitter

Slide 57

Slide 57 text

compute is a fungible commodity

Slide 58

Slide 58 text

emphasis on productivity

Slide 59

Slide 59 text

you can get a lot of awesome

Slide 60

Slide 60 text

dive in

Slide 61

Slide 61 text

just a little

Slide 62

Slide 62 text

Simple Storage Service S3

Slide 63

Slide 63 text

highly durable

Slide 64

Slide 64 text

99.999999999%

Slide 65

Slide 65 text

Highly scalable

Slide 66

Slide 66 text

Elastic Compute Cloud EC2

Slide 67

Slide 67 text

dynamic

Slide 68

Slide 68 text

autoscaling

Slide 69

Slide 69 text

EC2 instance types

Slide 70

Slide 70 text

standard “m1” high cpu “c1” high memory “m2” http://aws.amazon.com/ec2/instance-types/ EC2 instance types

Slide 71

Slide 71 text

text cluster compute instances

Slide 72

Slide 72 text

cluster GPU instances

Slide 73

Slide 73 text

http://aws.amazon.com/ec2/instance-types/ cluster compute “cc1” EC2 instance types cluster GPU “cg1”

Slide 74

Slide 74 text

10gbps

Slide 75

Slide 75 text

Placement Group

Slide 76

Slide 76 text

full bisection bandwidth

Slide 77

Slide 77 text

Linpack benchmark 880-instance CC1 cluster Performance: 41.82 TFlops* *#231 in Nov 2010 Top 500 rankings

Slide 78

Slide 78 text

KS for huge system at 1 k-point VERY DEMANDING network performance H size 56,000 (25GB) Runtime (16x8 processors) Local (Infiniband) 3h:48 Cloud (10Gbps) 1h:30 ($40) WIEN2k Parallel Performance Credit: K. Jorissen, F. D. Villa, and J. J. Rehr (U. Washington) •1200 atom unit cell; SCALAPACK+MPI diagonalization, matrix size 50k-100k

Slide 79

Slide 79 text

cost and use models

Slide 80

Slide 80 text

% Utilization time Ideal Effective Utilization Reserved Utilization On Demand Utilization Spot Utilization

Slide 81

Slide 81 text

making things easier

Slide 82

Slide 82 text

No content

Slide 83

Slide 83 text

No content

Slide 84

Slide 84 text

Elastic Beanstalk

Slide 85

Slide 85 text

No content

Slide 86

Slide 86 text

Heroku

Slide 87

Slide 87 text

% Utilization time Ideal Effective Utilization Reserved Utilization On Demand Utilization Spot Utilization

Slide 88

Slide 88 text

data at scale

Slide 89

Slide 89 text

some practical considerations

Slide 90

Slide 90 text

everything fails all the time

Slide 91

Slide 91 text

compute needs vary

Slide 92

Slide 92 text

new data/compute paradigms

Slide 93

Slide 93 text

No content

Slide 94

Slide 94 text

Amazon Elastic MapReduce

Slide 95

Slide 95 text

No content

Slide 96

Slide 96 text

doing stuff

Slide 97

Slide 97 text

No content

Slide 98

Slide 98 text

No content

Slide 99

Slide 99 text

No content

Slide 100

Slide 100 text

Customer Case Study: cyclopic energy http://aws.amazon.com/solutions/case-studies/cyclopic-energy/ OpenFOAM®

Slide 101

Slide 101 text

NASA JPL

Slide 102

Slide 102 text

Credit: Angel Pizzaro, U. Penn

Slide 103

Slide 103 text

http://aws.amazon.com/solutions/case-studies/numerate/

Slide 104

Slide 104 text

Bioproximity http://aws.amazon.com/solutions/case-studies/bioproximity/

Slide 105

Slide 105 text

No content

Slide 106

Slide 106 text

http://usegalaxy.org/cloud

Slide 107

Slide 107 text

mapreduce for genomics http://bowtie-bio.sourceforge.net/crossbow/index.shtml http://contrail-bio.sourceforge.net http://bowtie-bio.sourceforge.net/myrna/index.shtml

Slide 108

Slide 108 text

http://cloudbiolinux.org/

Slide 109

Slide 109 text

in summary

Slide 110

Slide 110 text

large scale data requires a rethink

Slide 111

Slide 111 text

data architecture

Slide 112

Slide 112 text

compute architecture

Slide 113

Slide 113 text

in infrastructure

Slide 114

Slide 114 text

the cloud

Slide 115

Slide 115 text

distributed, programmable infrastructure

Slide 116

Slide 116 text

rapid, massive, scaling

Slide 117

Slide 117 text

architecture evolved with the internet

Slide 118

Slide 118 text

can we build data science platforms?

Slide 119

Slide 119 text

there is no magic there is only awesome

Slide 120

Slide 120 text

two more things

Slide 121

Slide 121 text

10 minutes

Slide 122

Slide 122 text

http://aws.amazon.com/about-aws/build-a-cluster-in-under-10/

Slide 123

Slide 123 text

http://aws.amazon.com/education

Slide 124

Slide 124 text

[email protected] Twitter:@mndoci http://slideshare.net/mndoci http://mndoci.com Inspiration and ideas from Matt Wood& Larry Lessig Credit” Oberazzi under a CC-BY-NC-SA license