The steps required to address this issue of increased rate of progress are encapsulated in eight specific recommendations, which are summarized below.
RECOMMENDATION-1: Recognizing that many of the scientifically driven needs of computational chemistry can only be met by future generations of MPP computers, software and algorithm development should take both problem and resource scalability as a basic tenant.
RECOMMENDATION-2: Scalable parallel computational chemistry applications demand both advanced algorithm development and vendor-based hardware and software capabilities, targeted to the requirements of the computational chemistry community. A well-defined list of requirements must be tabulated, including those of portability, performance, enhanced development environments, scalable MPP subsystems (communication, disk I/O, memory capacity), compact expression of chemical algorithms, and available parallelism, including non-uniform memory access and multiple levels of granularity.
RECOMMENDATION-3: The development of new methods and algorithms should be encouraged and supported. A key aspect of this support should involve the integration of the computer science, computational science, and computational chemistry fields at the undergraduate and postgraduate levels.
RECOMMENDATION-4: Close interaction of computational chemistry application developers across chemistry disciplines and with computational scientists, computer scientists, and vendors should be strongly encouraged to foster co-education of these varied communities.
RECOMMENDATION-5: Development and production use of MPP supercomputers are necessary to make full use of today's MPP systems, while preparing for the next generation of computing resources. MPP supercomputer centers must be configured to meet the needs of computational chemistry applications and users.
RECOMMENDATION-6: MPP supercomputer centers should act as a repository for algorithms and expertise, and provide a focal point for communication between the computational communities, vendors, and application users.
RECOMMENDATION-7: Technology transfer to industrial researchers must be streamlined to increase the productive use of MPP-based computational chemistry applications.
RECOMMENDATION-8: A suite of computational chemistry benchmarks (c.f., the NAS kernels) needs to be developed to demonstrate requirements for comparison and development of software environments and hardware, as well as to provide performance measurements of MPP and conventional supercomputers.