With reference to numerical iterative algorithms, this paperexemplifies a methodology to design the runtime support of applicationssharing a set of data structures on a distributed memory architecture.According to the methodology, the support is decomposed into two layers:an application independent one, supplying the basic functionalities toaccess a shared structure, and an application dependent layer thatimplements the caching and prefetching strategies most appropriate forthe considered application. Starting from this assumption, we introduceDVSA, a package that implements the application independent layer andSHOB, one of the packages that can be developed on top of DVSA. SHOBdefines a weak consistency memory model where the user controls theamount of inconsistency due to caching and prefetching. The model iswell suitable to implement iterative numerical algorithms. Experimentalresults of the methodology are presented in the case of a uniformmulti-grid method to solve partial differential equations
展开▼
