1. Volunteer Computing: SETI and Beyond David P
Volunteer Computing: SETI and Beyond David P. Anderson University of California, Berkeley 7 June 2007

2. What is “Volunteer computing”?
Volunteers
Projects
Internet
Helps science
Involves public in science

3. Volunteer computing != Grid computing
Resource
owners
anonymous, unaccountable;
need to check results
identified, accountable
Managed
systems?
yes – software stack
requirements OK
no – need plug & play software
Clients behind
firewall?
yes – pull model
no – push model
ISP bill?
yes no
... nor is it “peer-to-peer computing”

4. Volunteer computing software
Application
Do scientific computation
Infrastructure
Manage work
Transfer files
Security
Screensaver
Accounting
BOINC
(Berkeley Open Infrastructure for Network Computing)

5. Creating a BOINC project
Get a server (Linux; ~$5K)
BOINC-enable application
Compile for various platforms
Develop programs to generate and handle work
Test
Create web pages about your research
Publicize
Keep volunteers informed

6. BOINC (Volunteer’s view)
1-click install, zero configuration
All platforms
Invisible, autonomic

7. Communication: “Pull” model
I’m a Windows/x86 computer with
512 MB RAM
20GB free disk
2.5 GFLOPS CPU
client
scheduler
Here are three jobs.
Job 1 has application files A,B,C,
input files C,D,E and output file F
...

8. Job replication Problem: can’t trust volunteers computational result
claimed credit
Application-specific checks, no replication
Replicated computing
do N copies, require that M of them agree
not bulletproof (collusion)
time
created validate; assimilate
Job x x x
created sent success
Instance x x x
created sent error
Instance x x x
created sent success
Instance x x x
created sent success
Instance x x x

9. How to compare results? Problem: numerical discrepancies
Stable problems: fuzzy comparison
Unstable problems
Eliminate discrepancies
compiler/flags/libraries
Homogeneous replication
send instances only to numerically equivalent hosts (equivalence may depend on app)

10. What applications can use BOINC?
Should be resource-intensive
1 CPU year: do it yourself
10,000 CPU years: use BOINC
Lots of independent tasks
Low data/compute ratio
Examples:
- Physical simulations (molecule, Earth, universe)
- CPU-intensive data analysis
- Search of large spaces (math)

11. Example: ClimatePrediction.net
Application: UK Met Office Unified Model
State-of-the-art global climate model
1 million lines of FORTRAN
High-dimensional search space
model parameters
boundary conditions
perturbed initial conditions

12. ClimatePrediction.net Using supercomputers: 1 day per run
10-20 total runs
Using BOINC:
6 months per run
50,000 active hosts
171,343 runs completed
Nature papers
60-fold savings

13. Other BOINC-based projects
LIGO; gravitational wave astronomy
U. Washington; protein study
U.C. Berkeley; SETI
CERN; accelerator simulation
STI, U. of Geneva; malaria epidemiology
IBM World Community Grid
several biomedical applications
...and about 30 others

14. Computing power Folding@home: 650 TeraFLOPS
200 from PCs; 50 from GPUs; 400 from PS3
BOINC-based projects:

15. The hard problems How to increase the number of volunteers?
currently 1 in 1000 PC owners
How to increase the number of projects?
currently stuck at about 50
How to get volunteers to diversify?

16. How to attract and retain volunteers?
Active hosts:
Retention
reminder s
frequent science updates
Recruitment
Software-based
“ a friend” mechanism
credit-based pyramid scheme?
Organizational
World Community Grid: “partner” program
Media coverage
need more discoveries
Commercial (bundling)

17. Why aren’t there more projects?
Lack of PR among scientists
IT antipathy
Creating a BOINC project is expensive:
Research group
Science
App development
Experiment design
Paper writing
Software/IT
Port/debug apps
workflow tools
server admin
Communications
Web site development
message board admin
public relations

18. Meta-projects Virtual Campus Supercomputing Center
Deployment and publicity:
PC labs, staff/faculty desktops
students
alumni
public
Research groups
Existing UCB staff
Science
App development
Experiment design
Paper writing
Software/IT
Port/debug apps
workflow tools
server admin
Communications
Web site development
message board admin
public relations
IBM World Community Grid

19. What else can volunteers do?
Testing
Translation
Program optimization
Message-board moderation
Online customer support
Skilled tasks

20. Conclusion Volunteer computing new paradigm for scientific computing
more/cheaper computing power
public involvement in science
enabling technology: BOINC
Talk to or me about:
Using BOINC for your research
Organizational use of BOINC
Other uses of volunteers to help research