Structure research in a sparse directory co-located with last-level cache

Directory is adopted by most CC-NUMA direct connection multiprocessor system. As a result, how to implement the directory is one of the key issues for the system performance. In this paper, we propose a novel technology, which divides up some space from the last-level cache (e.g., tertiary cache, L3) for storing the directory information. The directory space size could be adjusted flexibly according to the characteristics of different applications. What´s more, this approach offers high-level parallelism while suffering very little cost in the complexity of design, and almost has no effect on cache hit rate, when set properly.