HyperPascal for architecture independent parallel programming

Abstract

The design and implementation of a high-level parallel programming language interface for distributedm e m o r y M I M D architectures is described. This interface is termed as Hyperpascal and is based on a Computation Graph Decomposition model of parallel programming. Hyperpascal closely resembles the Pascal language but in addition provides constructs that allow for the high-level specification of process parallelism. Hyperpascal provides a machine-andependent view of parallel programming, simplifies parallel program design and development and facilitates portability of parallel programs. The current version of Hyperpascal has been implemented on the Intel iPSC/JSO Hypercube.

Building Grid MPI applications from modular components

“…The role of each process (atm or ocn) spawned would be determined by its rank by if (MyRank<=LastAtm) then atm else ocn; as the main program, specifying that processes with ranks 0 to LastAtm behave like atms and the rest like ocns. ocnP [1,1] atmP [2,6] atmP [2,8] atmP [2,7] atmP [3,7] atmP [3,6] atmP [3,8] …”

“…ocn or land) via its single port Down [1]. The actual value of context and port will depend on the application configuration.…”

“…For example it would be possible to print tracing lines such as ocn [1,1] sends to atm [2,6] via port Up [1] within context of Vertical [3] …”

“…Composed programs are pure MPI programs; we do not use any external environment for gluing components together. Other approaches dealing with a broader problem, that of the composition of heterogeneous components, develop component architectures (CCA [15], Charisma [1]). They manage "componentized" programs (mainly using OO techniques) as well as their communication.…”