Slide 1

Slide 1 text

Platforms for Deepak Singh P r i n c i p a l P r o d u c t M a n a g e r Data Science

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

bioinformatics

Slide 5

Slide 5 text

collection

Slide 6

Slide 6 text

curation

Slide 7

Slide 7 text

analysis

Slide 8

Slide 8 text

so what?

Slide 9

Slide 9 text

Image: Yael Fitzpatrick (AAAS)

Slide 10

Slide 10 text

lots of data

Slide 11

Slide 11 text

lots of people

Slide 12

Slide 12 text

lots of places

Slide 13

Slide 13 text

constant change

Slide 14

Slide 14 text

we want to make our data more effective

Slide 15

Slide 15 text

versioning

Slide 16

Slide 16 text

provenance

Slide 17

Slide 17 text

filter Via asklar under a CC-BY license

Slide 18

Slide 18 text

aggregate Image: Chris Heiler

Slide 19

Slide 19 text

extend Image: Bethan

Slide 20

Slide 20 text

human interfaces Image: Sebastian Anthony

Slide 21

Slide 21 text

share

Slide 22

Slide 22 text

image: Leo Reynolds communicate

Slide 23

Slide 23 text

hard problem

Slide 24

Slide 24 text

really hard problem

Slide 25

Slide 25 text

so how do get there?

Slide 26

Slide 26 text

information platforms

Slide 27

Slide 27 text

Image: Drew Conway

Slide 28

Slide 28 text

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

Slide 29

Slide 29 text

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

Slide 30

Slide 30 text

accept all data formats

Slide 31

Slide 31 text

evolve APIs

Slide 32

Slide 32 text

data as a programmable resource

Slide 33

Slide 33 text

data is a royal garden

Slide 34

Slide 34 text

compute is a fungible commodity

Slide 35

Slide 35 text

constraints everywhere

Slide 36

Slide 36 text

Hardware CPU, storage, memory Data management Collections, datasets, provenance Software parallelization, optimization Availability Backup, redundant, replicated Cost Small

Slide 37

Slide 37 text

remove constraints Credit: Pieter Musterd a CC-BY-NC-ND license

Slide 38

Slide 38 text

amazon web services

Slide 39

Slide 39 text

your infrastructure

Slide 40

Slide 40 text

No content

Slide 41

Slide 41 text

ec2-run-instances

Slide 42

Slide 42 text

No content

Slide 43

Slide 43 text

secure global on demand

Slide 44

Slide 44 text

programmable

Slide 45

Slide 45 text

No content

Slide 46

Slide 46 text

No content

Slide 47

Slide 47 text

No content

Slide 48

Slide 48 text

elastic

Slide 49

Slide 49 text

Netflix needed to transcode 17,000 titles (80TB of data) to support the launch of Sony PS3. They provisioned 1200 Amazon EC2 instances and completed the transcoding process in just days. Source: Adrian Cockroft (Netflix)

Slide 50

Slide 50 text

Source: Adrian Cockroft (Netflix)

Slide 51

Slide 51 text

durable

Slide 52

Slide 52 text

99.999999999%

Slide 53

Slide 53 text

I did say data was a royal garden

Slide 54

Slide 54 text

performance

Slide 55

Slide 55 text

“Our 40-instance (m2.2xlarge) cluster can scan, filter, and aggregate 1 billion rows in 950 milliseconds.” Mike Driscoll - Metamarkets

Slide 56

Slide 56 text

WIEN2K Parallel Performance H size 56,000 (25GB) Runtime (16x8 processors) Local (Infiniband) 3h:48 Cloud (10Gbps) 1h:30 ($40) 1200 atom unit cell; SCALAPACK+MPI diagonalization, matrix size 50k-100k Credit: K. Jorissen, F. D. Villa, and J. J. Rehr (U. Washington)

Slide 57

Slide 57 text

“Our tests have shown more than 90 percent scaling efficiency on clusters of up to 128 GPUs each”

Slide 58

Slide 58 text

consumption models

Slide 59

Slide 59 text

on-demand

Slide 60

Slide 60 text

Reserved Instances

Slide 61

Slide 61 text

what is the value of your data?

Slide 62

Slide 62 text

No content

Slide 63

Slide 63 text

No content

Slide 64

Slide 64 text

the clouds biggest value

Slide 65

Slide 65 text

remove constraints

Slide 66

Slide 66 text

No content

Slide 67

Slide 67 text

Image: Chris Dagdigian

Slide 68

Slide 68 text

Credit: Angel Pizzaro, U. Penn

Slide 69

Slide 69 text

13k sequences - 10 min - 0.1s per sequence

Slide 70

Slide 70 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 71

Slide 71 text

No content

Slide 72

Slide 72 text

No content

Slide 73

Slide 73 text

No content

Slide 74

Slide 74 text

30,472 cores

Slide 75

Slide 75 text

$1279/hr

Slide 76

Slide 76 text

http://cloudbiolinux.org/

Slide 77

Slide 77 text

http://usegalaxy.org/cloud

Slide 78

Slide 78 text

“The process of moving StarMolsim over to the cloud to support the “Introduction to Modeling and Simulation” course at MIT was a huge success. The cloud enabled the STAR group to move away from the responsibility of owning and maintaing dedicated hardware and instead focus on their core mission of developing software and services for faculty, students, and researchers at MIT” http://web.mit.edu/stardev/cluster/about.html

Slide 79

Slide 79 text

in summary

Slide 80

Slide 80 text

large scale data requires a rethink

Slide 81

Slide 81 text

data architecture

Slide 82

Slide 82 text

compute architecture

Slide 83

Slide 83 text

distributed, programmable infrastructure

Slide 84

Slide 84 text

amazon web services

Slide 85

Slide 85 text

remove constraints

Slide 86

Slide 86 text

can we build data science platforms?

Slide 87

Slide 87 text

there is no magic there is only awesome

Slide 88

Slide 88 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