High Level Language Memory Management on Parallel Architectures.

摘要

HEP memory management packages such as YBOS and ZEBRA have been implemented and are currently running on a variety of mainframe computers. These packages were originally designed to run on single CPU engines. Implementation of these packages on parallel machines, loosely or tightly coupled architectures is discussed. ZEBRA (CERN package) on ACP (Fermilab) is presented in detail. Design of memory management system for the new generation of ACP systems or similar parallel architectures are presented. The future of packages such as ZEBRA is not only linked to system architecture, but also to languages issues. We briefly mention penalties in using F77 with respect to other increasingly popular languages in HEP, such as C, on parallel systems. 9 refs.