Extend the concept of shared memory to abstract data type.
Design and implement a selection of shared ADT library functions.
Evaluate alternative distribution, replication and scheduling techniques using simulation and analytical modelling.
Demonstrate use of shared ADTs in example applications codes running under simulation and on available parallel equipment.
Redesigned and reimplemented "future-space" ADT for Fujitsu AP1000.
A pilot implementation of this ADT was described in the grant proposal. A redesigned implementation has been constructed and further research continues on its application in transaction processing. This can naturally be extended to use a memoisation table and to use a branch-and-bound mechanism for search applications.
Designed and partially implemented adaptive mesh refinement ADT.
This simple adaptive mesh refinement (AMR) package manages a hierarchy of patched-grids suitable for "h-refinement" in 2-D applications using non-conforming patched grids.
The ADT handles computation over original or derived patches as well as boundary exchange, refinement and contraction, and will manage load balancing, process placement and data movement. Performance evaluation for cache-coherency and communication optimisations with alternative distribution, replication and scheduling techniques on both shared-memory and distributed-memory systems will be made using simulation tools.