How did they do that - BOINC!
I see that IBM and Sun are renting out supercomputer time (IBM on Blue Gene). $1/hour per processor used at Sun. A few years ago, I read an article about an oil company whose administrative and engineering desktop computers ran linux. At night, they rebooted to become a single beowulf cluster. Cool way to get double usage out of their hardware, no? Now I see there is a cottage industry for supercomputing based on beowulf, and I also see that Microsoft is getting into the act. Applications include analyzing sesimic data for oil deposits, atmosphere models for weather and climate prediction, genome data, and chemical compound data for pharmaceuticals.
But, I have to ask - is building a new supercomputer or renting time on one efficient?
Wouldn't it be cheaper to pay home users, say, $0.10/hour for them to participate in a distributed computing project? The infrastructure is already there: it's called BOINC (for Berkeley Open Infrastructure for Network Computing). There are several projects running on BOINC, of which the Search for Extra Terrestrial Intelligence (SETI) is the most famous.
There are two barriers I can identify. One is that the BOINC program needs a billing mechanism, but that should be no problem. The other is that when you are sending out proprietary data to project participants, what keeps your competitor from participating and intercepting your data? That problem seems to be bigger for pharmaceutical companies than for oil explorers, who can after all hide the location of their data. Pharmaceuticals may be able to hide data (I don't know enough about it to know what is or isn't possible) or they may be able to insert misinformation. On the other hand, you can always screen the participants and sign them up to nondisclosure agreements, but that defeats one of the advantages of distributed computing, which is that you have low transaction costs so you can keep the volunteers coming.
What other public goods projects (like climate and SETI) could benefit from the distributed approach? Are there other private goods that could be produced this way? I know that Bill Gates proposed a while back that we consider adopting a system where every e-mail you send commits you to a certain amount of CPU time for such projects (say, a minute). That would be no burden to most of us because it could run in the background, but it would shut a spammer down. In a way, you would essentially be calculating the solution to a public good: the spam-free internet (it wouldn't matter if the calculations were just multiplication tables). If you also happened to produce a cure for Alzheimer's, that would be good, too.
Now that I've posed the question, I have to ask another: is $0.10 enough to induce participants? Or, is it too much because it will encourage people to use more electricity and therefore create more pollution? We may already have too much electricity wasted on electronics, won't this create more?
But, I have to ask - is building a new supercomputer or renting time on one efficient?
Wouldn't it be cheaper to pay home users, say, $0.10/hour for them to participate in a distributed computing project? The infrastructure is already there: it's called BOINC (for Berkeley Open Infrastructure for Network Computing). There are several projects running on BOINC, of which the Search for Extra Terrestrial Intelligence (SETI) is the most famous.
There are two barriers I can identify. One is that the BOINC program needs a billing mechanism, but that should be no problem. The other is that when you are sending out proprietary data to project participants, what keeps your competitor from participating and intercepting your data? That problem seems to be bigger for pharmaceutical companies than for oil explorers, who can after all hide the location of their data. Pharmaceuticals may be able to hide data (I don't know enough about it to know what is or isn't possible) or they may be able to insert misinformation. On the other hand, you can always screen the participants and sign them up to nondisclosure agreements, but that defeats one of the advantages of distributed computing, which is that you have low transaction costs so you can keep the volunteers coming.
What other public goods projects (like climate and SETI) could benefit from the distributed approach? Are there other private goods that could be produced this way? I know that Bill Gates proposed a while back that we consider adopting a system where every e-mail you send commits you to a certain amount of CPU time for such projects (say, a minute). That would be no burden to most of us because it could run in the background, but it would shut a spammer down. In a way, you would essentially be calculating the solution to a public good: the spam-free internet (it wouldn't matter if the calculations were just multiplication tables). If you also happened to produce a cure for Alzheimer's, that would be good, too.
Now that I've posed the question, I have to ask another: is $0.10 enough to induce participants? Or, is it too much because it will encourage people to use more electricity and therefore create more pollution? We may already have too much electricity wasted on electronics, won't this create more?
Labels: science
<< Home