Abstract
Recent changes in hardware architecture have emphasized the need for suitable programming methodology for parallel or parallelizable applications. Besides avoiding parallelism-related errors, such a system must be able to employ the computing power by reorganization of both control flow and data structures.
Current parallel-programming systems either leave a major burden on the programmer (like C++/MPI) or limit the strength of the language in order to allow automatic parallelization (like FORTRAN). In reality, there are many degrees of programmer's awareness of parallelization; in the middle of this spectrum, the necessary means are still missing.
In this paper, the requirements for the hypothetical programming language or system are collected and several options to adapt existing means to these requirements are compared.