(15) An Implementation of a Parallel Active Database System in KLIC

To enhance the function and performance of database systems, we are now implementing a parallel active database engine, named Parade, by using KLIC. An active database system, an integration of a DBMS and a rule base system, is very useful for many database applications, but has a vulnerable point in its performance. We have proposed several methods for speeding up the active database system by assuming distributed memory parallel environment. Inter-database-query parallelism is realized by nested transaction using perpetual processes of KLIC. Pipeline execution with a shared logical variable of KLIC is suited for inter-database-operation parallelism. Intra-database-operation parallelism is realized by data parallel execution with horizontal data distribution. For rule operations, a sub-transaction of the nested transaction framework is assigned to a rule activation as a coarse grain parallel execution. Conditions in the rules are checked by making a lot of tokens simultaneously traverse discrimination networks with hash functions. It is fine grain parallelism of the rule execution. Its performance can further be improved by using processor clusters with software cache for tokens. Moreover, time for optimizing the discrimination networks is hidden by parallel execution. Since a naive parallel system decreases the system reliability in spite of the durability is another important issue of databases, we have also proposed a fault tolerant parallel software mechanism to keep the reliability of Parade high.