wiki:WhyUseBoinc

Version 2 (modified by davea, 16 years ago) (diff)

--

Why use BOINC?

Use BOINC if you need lots of cheap computing power. BOINC lets you access two classes of computing resources:

Cost comparison

Suppose you need a lot of computing power - say, 100 TeraFLOPS for 1 year. Here are some ways you can get it:

Use Amazon's Elastic Computing Cloud : $175 Million
Based on $0.10 per node/hour.
Build a cluster: $12.4 Million
This includes power and air-conditioning infrastructure, network hardware, computing hardware, storage, electriciy, and sysadmin personnel.
Use BOINC: $125,000
based on the average throughput and budget of the 6 largest volunteer computing projects.

What's the catch?

To get lots of computing power this way, you'll need to do some unusual things:

  • Publicize and promote your project, on the web and if possible in the media.
  • Devote some resources to managing and communicating with your volunteer community.
  • Get your applications to run on a wide range of computer types.

What applications can BOINC handle?

BOINC works best with "bag of task" computations - large sets of independent jobs - with modest memory and storage requirements. Typical applications include:

  • Simulations of physical systems
  • Compute-intensive analysis of large data sets
  • Exploration of large search spaces (including genetic algorithm).

BOINC doesn't require you to rewrite your applications. You can even use existing executables without source code.

Organizational possibilities

If you're part of an organization such as a university, you should consider the various organizational levels at which BOINC can be used. There is an overhead in creating and maintaining a BOINC project, and it can be advantageous to amortize this overhead across multiple scientists. Some possibilities:

Individual
Some projects (Primegrid, chess960) are run by private individuals.
Scientist
Most current projects are run by individual scientists or research groups.
Consortium
Einstein@home is run by a multi-university research consortium (LIGO).
Application-centered academic community
Mindmodeling.org serves researchers from about 20 universities who all use the same application (the ACT-R cognitive modeling system).
Research institute
LHC@home servers multiple groups at CERN
Umbrella organization
IBM World Community Grid handles applications from ~8 different research groups.
Campus-level project
This would provide a Virtual Campus Supercomputing Center; there are no examples yet.