An Extended Master Worker Model for a Desktop Grid Computing Platform (QADPZ)

In this paper we first present briefly QADPZ, an open source platform for heterogeneous desktop grid computing, which enables users from a local network (organization-wide) or Internet (volunteer computing) to share their resources. Users of the system can submit compute-intensive applications to the system, which are then automatically scheduled for execution. The scheduling is made based on the hardware and software requirements of the application. Users can later monitor and control the execution of the applications. Each application consists of one or more tasks. Applications can be independent, when the composing tasks do not require any interaction, or parallel, when the tasks communicate with each other during the computation. QADPZ uses a master worker-model that is improved with some refined capabilities: push of work units, pipelining, sending more work-units at a time, adaptive number of workers, adaptive timeout interval for work units, and use of multithreading, to be presented further in this paper. These improvements are meant to increase the performance and efficiency of such applications.