ABSTRACT
This chapter is a chapter about the practical issues of using a particular Desktop Grid middleware, namely PastryGrid [ACJ08, ACJM10]. This middleware differs from other popular Desktop Grid middleware in the sense that it is built with the Peer-to-Peer philosophy in mind. The implementation is done on top of a popular P2P system, namely Pastry, and hence its name. PastryGrid is able to execute task graphs with precedences in a fully distributed manner. During the execution of the task graph, the choice of the machines that will execute the next tasks is unknown but it is set on the fly, after the completion of each task. This feature relies on a concept of Rendez-vous point (RDV) and we manage one RDV per application. To be clear, the PastryGrid overlay network is able to manage multiple applications at the same time. The most important fact for a machine is to join the overlay network. After that,
a machine will become a worker for another machine, or a RDV point or will play yet another role for mastering the fault tolerance of the application.
