SYMPOSIUM MINISYMPOSIA

ABSTRACT

The Master-Worker is a classic parallel programming model. One processor (master) is devoted to assigning computations to all the other processors (workers) involved. When a worker processor completes a task, a request for more work is sent to the master processor, which replies with one of two messages: a message containing the next task to be computed, or a message signaling the job has been completed if all work has been done. To execute this programming model, the use of a message passing library (e.g., MPI or PVM) is essential.

While the Master-Worker model is easily programmed to run on a single parallel platform, running such a model for a single application across distributed machines presents interesting challenges. In order to meet those challenges, we have developed the Queen-Drone programming model for distributed computation. This name was chosen for the analogy of a single queen bee assigning jobs to drone bees; disjoint groups of drones are collected into hives (corresponding to groups of parallel worker processors executing on separate parallel machines). We have programmed this model using MPI_Connect.

MPI_Connect is a metacomputing middleware that allows separately initiated MPI applications to interact in a peer-peer fashion. The MPI codes interoperate by naming themselves with a third party (transparent) naming service that coordinates the interaction. The names used are plain ASCII and are completely freeform to aid debugging. Once MPI applications have named themselves they can then locate other MPI applications by using this naming service. When both parties agree, an MPI intercommunicator is created between them and this is then used in the same way as any other intercommunicator in MPI. Initial versions of the system used PVM and were thus called PVMPI. The later versions, called MPI_Connect, used either PVM or a new metacomputing system called SNIPE, also from the University of Tennessee.

After describing our algorithm, we present some experiences and execution results of the Queen-Drone model applied to a simple integer factoring code and a large-scale harbor wave response code. We have been able to run experiments on machines of the same architecture, as well as between machines of different manufacture.

Getting More Information

To obtain more information about the meeting send e-mail to: fscc98@pax.st.usm.edu.

Return to PAX
Return to USM

USM
inquire@delphi.st.usm.edu
Updated
http://pax.st.usm.edu