In order to be truly portable, the building blocks underlying parallel software libraries must be standardized. The definition of computational and message-passing standards [14][12] provides vendors with a clearly defined base set of routines that they can optimize. From the user's point of view, standards ensure portability. As new machines are developed, they may simply be added to the network, supplying cycles as appropriate.
From the mathematical software developer's point of view, portability may require significant effort. Standards permit the effort of developing and maintaining bodies of mathematical software to be leveraged over as many different computer systems as possible. Given the diversity of parallel architectures, portability is attainable to only a limited degree, but machine dependences can at least be isolated.
Scalability demands that a program be reasonably effective over a wide range of numbers of processors. The scalability of parallel algorithms over a range of architectures and numbers of processors requires that the granularity of computation be adjustable. To accomplish this, we use block algorithms with adjustable block sizes. Eventually, however, polyalgorithms (where the actual algorithm is selected at runtime depending on input data and machine parameters) may be required.
Scalable parallel architectures of the future are likely to use physically distributed memory. In the longer term, progress in hardware development, operating systems, languages, compilers, and communication systems may make it possible for users to view such distributed architectures (without significant loss of efficiency) as having a shared memory with a global address space. For the near term, however, the distributed nature of the underlying hardware will continue to be visible at the programming level; therefore, efficient procedures for explicit communication will continue to be necessary. Given this fact, standards for basic message passing (send/receive), as well as higher-level communication constructs (global summation, broadcast, etc.), are essential to the development of portable scalable libraries. In addition to standardizing general communication primitives, it may also be advantageous to establish standards for problem-specific constructs in commonly occurring areas such as linear algebra.