Hiroyasu Nishiyama

Pseudo-vectorizing Compiler for the SR8000

The pseudo-vector processing (PVP) is a framework that enables fast processing similar to vector processing. In this paper, we describe the compiler optimizations that effectively utilize PVP on the SR8000. These include access method analysis, preloading, and prefetching optimizations. Evaluations on the SR8000 indicate that PVP can effectively hides memory latency.

Fast synchronization for quasi‐immutable objects

Increasing synchronization performance is one of the most important factors for improving Java performance. We present the concept of quasi‐immutable objects and its application to implement fast synchronization primitives. Quasi‐immutable objects are usually used for read accesses, but they are rarely used for write accesses. Using these characteristics, we describe the optimization for a synchronized read sequence of quasi‐immutable objects. Our described method dynamically detects objects that are not written from critical sections of other threads, and applies the fast read sequence. The optimized read sequence enables the simultaneous execution of critical sections by multiple threads and reduces synchronization overhead. We implemented the described optimization on HotSpot Java VM. The experimental evaluations indicated that the optimization improved the performance of a micro‐benchmark up to a factor of 16 and SPEC benchmarks up to 31%. Copyright

Efficient Query Processing of Semantic Data Using Graph Contraction on RDBMS

Efficient store and query of RDF graph database is of increasing importance due to the popularity and widespread acceptance of RDF on various applications including semantic/situation aware computing. In this paper, we have applied an query optimization based on graph contraction that boosts the query processing on relationally-backed RDF store. The query optimization technique based on graph contraction creates summarized graph in advance and uses it to efficiently query the original dataset. We used decomposition storage model on top of RDBMS to represent RDF graph, and applied the optimization technique based on graph contraction. Preliminary experiments using MySQL and synthetic dataset showed that the method can improve the query performance by 6.62 times. This means that the contraction based optimization on RDBMS is promising technique for retrieving data in semantic/situation aware computing.