AMOS-SDDS: A Scalable Distributed Data Manager for Windows Multicomputers

Known parallel DBMS offer at present only static partitioning schemes. Adding a storage node is then a cumbersome operation that typically requires the manual data redistribution. We present an architecture termed AMOS-SDDS for a share-nothing multicomputer. We have coupled a high-performance main-memory DBMS AMOS-II and a manager of Scalable Distributed Data Structures (SDDS) into a scalable distributed system. SDDS provides the scalable data partitioning in distributed RAM, supporting parallel scans with function shipping. AMOS-SDDS couples both systems using essentially the AMOS-II foreign function interface. The scalability that appeared from our experiments abolishes the cumbersome storage limits of a single site RAM DBMS technology. Its RAM query processing and scalable data partitioning are an improvement over the current parallel DBMSs technology. We validate AMOS-SDDS architecture by experiments with distributed nested loop join queries over a file scaling up to 300.000 tuples. It includes performance study of speed-up and scale-up characteristics. The results encourage the use of SDDS for modern high-performance database systems.