Enabling rapid development of parallel tree search applications

摘要: Virtual observatories will give astronomers easy access to anunprecedented amount of data. Extracting scientific knowledge from these data increasingly demand both efficient algorithms as well the power parallel computers. Nearly all analyses large astronomical datasets use trees their fundamental structure. Writing tree-based techniques, a task that is time-consuming even on single-processor computers, exceedingly cumbersome massively platforms (MPPs). Most applications run MPPs are simulation codes, since expense developing them offset by fact they be used for many years researchers. In contrast, analysis codes change far more rapidly, often unique individual researchers, and therefore accommodate little reuse. Consequently, economics current high-performance computing development paradigm does not favor applications. We have built library, called Ntropy, provides flexible, extensible, easy-to-use way serial platforms. Our experience has shown only our library save time, it can also deliver excellent performance scalability. Furthermore, Ntropy makes an astronomer with or noparallel programming quickly scale application distributed multiprocessor environment. By minimizing time scalable analysis, we enable wide-scale discovery massive datasets.