A document engine on a db cluster

Requirements on document engines include low query response time and freshness of index data. Our solution to the problem is based on a DB cluster consisting of PCs, each of them running an o -the-shelf DBMS. The specialty of our approach is to allow concurrent execution of insertion and retrieval on the same data. Our technique is based on decomposing and parallelizing insertion and retrieval requests, while at the same time guaranteeing correctness by an additional high-level transaction manager. An important design decision with our architecture is how to assign data to the di erent components, and we compare several such alternatives. The speed-up obtained is surprisingly good. The cluster-based architecture nicely defuses the bottlenecks occurring with a single-component system.