This is a PhD thesis describing implementation of an object-oriented library for shared-memory parallel programming. Although completed in 1996, the work has become relevant again with the growth of commodity multicore processors. While some details of the programming style would differ with advances in C++ since the mid-1990s, the fundamentals of structuring code to match characteristics of the memory hierarchy still apply. Areas covered include caches and the address translation mechanism (specifically interactions with the TLB). The material is understandable for a reader with a reasonable understanding of computer architecture and C++ programming. Some algorithms background would also be useful. A few comments on technology changes since 1996 have been added to the text.