25
from imagination to impact © 2011 NICTA. All Rights Reserved. Maximilian Ott With lots of help from Thierry Rakotoarivelo & Guillaume Jourjon Will RSpec help us to do better Science?

Will RSpec help us to do better science

  • Upload
    maxott

  • View
    107

  • Download
    2

Embed Size (px)

DESCRIPTION

These are the slides for a plenary talk I gave at GEC 11 in Denver in July 2011. The primary message is that we, the networking community need to fundamentally improve the way we do experimental research.

Citation preview

Page 1: Will RSpec help us to do better science

from imagination to impact © 2011 NICTA. All Rights Reserved.

Maximilian Ott With lots of help from Thierry Rakotoarivelo & Guillaume Jourjon

Will RSpec help us to do better Science?

Page 2: Will RSpec help us to do better science

2 © 2011 NICTA

Page 3: Will RSpec help us to do better science

3 © 2011 NICTA

Why are we doing what we are doing?

Page 4: Will RSpec help us to do better science

4 © 2011 NICTA

Curious Bunch - Want to figure out new things

Page 5: Will RSpec help us to do better science

5 © 2011 NICTA

The advancement of science depends in general on the interplay between experimental studies and theory

Wikipedia

Page 6: Will RSpec help us to do better science

6 © 2011 NICTA

Many interesting Things only happen at Scale

Young Hyun, CAIDA

Page 7: Will RSpec help us to do better science

7 © 2011 NICTA

Need for LARGE Experimental Facilities

Page 8: Will RSpec help us to do better science

8 © 2011 NICTA

Page 9: Will RSpec help us to do better science

9 © 2011 NICTA

Page 10: Will RSpec help us to do better science

10 © 2011 NICTA

Hypothesis

Experiment

Observation

Analysis

Page 11: Will RSpec help us to do better science

11 © 2011 NICTA

Hypothesis

Experiment

Observation

Analysis

Page 12: Will RSpec help us to do better science

12 © 2011 NICTA

How “well” do we do experiments today?

Missing Model

Validation

Missing Experiment Description

Missing Replication Information

Missing Precision

Information

OK

30% (10)

“Evaluation” section of SIGCOMM 2010 papers

58% (19)

21% (7) 9%

(3) 12% (4)

Page 13: Will RSpec help us to do better science

13 © 2011 NICTA

How can we improve this?

•  The Art & Science of Experimentation –  The art of designing experiments –  Scientifically sound methodology & analysis

•  Dissemination –  Overcoming the 6 page paper information barrier

•  Learning –  Reflection on the methodology –  Developing Best Practices

•  Tools –  To streamline processes –  To record processes and outcomes

Page 14: Will RSpec help us to do better science

14 © 2011 NICTA

Need for Methodologies & Best Practices

Observations

Model 1 Model 2 …

Ha , H0 Ha , H0

Experiment Experiment

New Obs + Analysis

Page 15: Will RSpec help us to do better science

15 © 2011 NICTA

Need to teach Experimenting at Scale

Internet

High Speed

Dedicated

Network

Control

Network

Remote

Experiment

Server

Student

Student

Experiment

Queue

Tested application and protocol

Measurement points

Network

In the Box

ireel.npc.nicta.com.au

Page 16: Will RSpec help us to do better science

16 © 2011 NICTA

Need new ways to disseminate Findings

Authors are invited to submit Full Papers of up to 10 pages and Testbed Practices Papers of up to 6 pages, including references, figures and tables, formatted according to the IEEE conference format. ?

Page 17: Will RSpec help us to do better science

17 © 2011 NICTA

LabWiki: Publishing Your Lab Book

Studied Phenomenon

Model 1

AnalysesFalsify and/or

refine the models

HA , H0

ExperimentExperimentExperiment

Observations

Model i

HA , H0

ExperimentExperimentExperiment

...

Portal

OEDL Script

Measure-ments

database

R Script

0

20

40

60

80

100

120

140

160

180

200

0 50 100 150 200 250 300 0

50000

100000

150000

200000

250000

300000

350000

Dis

tan

ce

(m

)

Ba

nd

wid

th

Time (s)

DistanceBandwidth

Page 18: Will RSpec help us to do better science

18 © 2011 NICTA

Observations

Model 1 Model 2 …

Ha , H0 Ha , H0

Experiment Experiment

New Obs + Analysis

Notes

Code

Experiment Description

Analysis

Runs

Measurements

Page 19: Will RSpec help us to do better science

19 © 2011 NICTA

Observations

Model 1 Model 2 …

Ha , H0 Ha , H0

Experiment Experiment

New Obs + Analysis

Notes

Code

Experiment Description

Analysis

Runs

Measurements

Page 20: Will RSpec help us to do better science

20 © 2011 NICTA

Observations

Model 1 Model 2 …

Ha , H0 Ha , H0

Experiment Experiment

New Obs + Analysis

Notes

Code

Experiment Description

Analysis

Runs

Measurements

Summary

Page 21: Will RSpec help us to do better science

21 © 2011 NICTA

Summary

Researcher

Summary

Reader

Comment

Reviewer

Analysis

Review

Page 22: Will RSpec help us to do better science

22 © 2011 NICTA

We are working on it

•  OMF –  OEDL to describe experiments –  Experiment Controller to run them

•  OML –  Capture measurements & meta-data

•  Portal –  Wiki: Keep notes with references –  Schedule & monitor runs –  Analysis with integrated R server –  Repository for all artifacts

Page 23: Will RSpec help us to do better science

23 © 2011 NICTA

App 1

App 2

e1

e2

c1

Measure

ment Lib

rary

(O

ML)

ControlNetwork

. . .

. . .

Experiment Node

OML Server

SQL Database

Control Node

Proxy Server

ON/OFF

Sensingdaemon

System Monitor

sensor device

Y/Z

Coo

rdin

tate

X Coordinate

Link Quality:Very GoodGoodPoorVery Poor

Wiki

OMF Portal

OML

R OEDL Visualization

Page 24: Will RSpec help us to do better science

24 © 2011 NICTA

Call to Action

•  The Art & Science of Experimentation –  The art of designing experiments –  Scientifically sound methodology & analysis

•  Dissemination –  There need to be better ways than 10 page papers

•  Learning –  Reflection on the methodology –  Developing Best Practices –  Teaching experimentation at scale

•  Tools –  To streamline processes –  To record processes and outcomes

Page 25: Will RSpec help us to do better science

25 © 2011 NICTA

omf.mytestbed.net wiki.mytestbed.net [email protected]