DAME: An environment for preserving the efficiency of data-parallel computations on distributed systems

Data parallel programming is the most widely adopted paradigm for a large class of problems on traditional multicomputers (see SPMD programming model sidebar, p. 23). Nevertheless, it is a very hard task to preserve efficiency when this style is adopted on a cluster of heterogeneous nodes having nonuniform and time varying computational powers. Very popular packages, such as PVM and MPI, 1,2 allow the programmer to use these clusters as an homogeneous parallel virtual machine. However, they do not avoid the potential inefficiencies due to the unpredictable variability of usually shared resources. DAME, an acronym for DAta Migration Environment, overcomes such drawbacks by means of transparent supports that hide irregular network topology, adapt the data distribution dynamically to platform conditions, and mask the consequent nonuniform distribution to the programmer