Distributed Computing: It's OK To Share Data with Lots of People

Discussion on the separation between those who generate models and those who generate experimental data in Birnholtz and Bietz article this week led me to thinking about larger collaborative linkages in research. We could also consider the linkages between hardware designers, software designers, and the different demands and constraints placed on these groups. One public example that shows the collaboration between researchers generating experimental data and those who process it can be seen in the many distributed computing projects such as SETI@Home, Folding@Home, and others.

As a brief introduction, Folding@Home and projects like it rely on thousands, if not millions (SETI@Home has 3 million users currently) of users around the globe to download experimental data to their computer and process the data. In the case of SETI@Home, the data consists of radio telescope data that is analyzed for signs of extraterrestrial life, but other projects investigate biochemistry (protein folding) and many other scientific queries. Processing is usually automatic, but since users are downloading the data, the experimenters must trust that users will not tamper with the data in any way.

A posting from August 20 of this year on the Folding@Home blog mentioned the danger of tampering with data downloaded from the Folding@Home project. Users may tamper with data with completely benign intentions, such as maximizing their processor time, but we can see that data transformations might taint the overall data, and for this reason it is not allowed by this project. This is a simple example of something we have seen again and again: researchers need to maintain control over their data, especially in this case, where they are collecting results from processing done on their data with the future intention of making discoveries. This is a model for the way science will be done in the years to come: data is collected, stored, and then farmed out to processing teams, and fits in with the "stream" model in the aforementioned paper, instead of the discrete event data model. The fact that the data can be broken up for parallel processing enables this sort of massive crowdsourcing. Despite the huge gains in processor power over the past few decades, some problems are just too complex to solve overnight, no matter how many processor cores you have.

There is one additional benefit to projects like SETI@Home and Folding@Home: they provide an incentive to participate for users in the form of groups that can compete for prestige by contributing the largest number of processed hours, and there is a feeling that one is "helping out" to solve complex scientific problems. We have seen that one of the problems in getting scientists to share their data is figuring out how to tie the social and scientific rewards together, and I feel that these projects do this well.