Volunteer Cloud Computing: MapReduce over the Internet

Volunteer Computing harnesses computing resources of machines from around the world to perform distributed independent tasks, following a master/worker model. Despite the recent increase in popularity and power in middleware such as BOINC, there are still several limitations in existing systems. Current research is oriented towards optimizing existing applications, while the number of active users and projects has reached a plateau. A programming paradigm that has been significantly popular and is used by several systems on the cloud is MapReduce. The main advantage of this paradigm is that it can be used to solve a vast amount of different problems, by breaking them into simple steps and taking advantage of distributed resources. Volunteer Computing provides these resources, and although it cannot match the conditions offered by a cluster, it has other advantages that can be leveraged. In this paper, we try to increase the computational power of Volunteer Computing systems by allowing more complex applications and paradigms such as MapReduce to be run, thus opening new avenues and possibilities for the use of computational devices scattered through the Internet. We created a BOINC prototype that can run MapReduce jobs (BOINC-MR), using a pull-model in which communication is always initiated by the client. By running experiments on a small cluster, with multiple variables, we were able to evaluate a few initial scenarios with this paradigm. We used a simple MapReduce application, word count, as proof of concept, just to demonstrate a typical execution.