Problem: Really large scientific computations.
Is it real? No matter how fast computers get, we seem to find more computations to run on them.
Nugget: Distribute the computation to regular computer users through a client-side platform that runs when the computer is idle (e.g. as a screensaver) and incentivizes users by providing a nice UI and a count of how much they are contributing to the project. The paper describes a lot of neat features that went into BOINC to make it easy for scientists and attractive for users of various levels (enthusiasts vs. average coputer users).
Influence: SETI@home is very popular (although no aliens have been found yet!) and other projects also seem to have made a significant amount of progress using BOINC. It's a good idea. The lessons learned with respect to building a system that real computer users will want to participate in are also great.
Criticism: Interestingly, BOINC is perhaps the least power-efficient way possible to get really massive parallel computations (compared to a data center or to a supercomputer).
1 comment:
Excellent comment about the power efficiency! On the other hand, in many cases home computers are on 24x7 whether or not they run BOINC.
Anyway, it would be interesting to quantify the power-efficiency (inefficiency) of a system such as BOINC vs. datacenters.
Post a Comment