Spirit-III: an advanced relational database machine introducing a novel data-staging architecture with Tuple Stream Filters to preprocess relational algebra

• Authors: Noriyuki Kamibayashi;Kazuo Seo
• Venue: AFIPS '82 Proceedings of the June 7-10, 1982, national computer conference
• Year: 1982

This paper proposes an advanced architecture of the relational database machine (RDBM), named SPIRIT-III, which is basically organized into a three-level memory hierarchy with a sophisticated data-staging and preprocessing architecture for executing relational algebra. SPIRIT-III aims at totally improving both I/O and CPU processing boundary problems and has two major architectural features. One is the introduction of the relational-database-oriented data-staging mechanism, called the look-ahead data-staging mechanism, which can optimally schedule data movement in the memory hierarchy. The other is to attach refined preprocessing mechanisms for relational algebra operations to data transfer lines connected between each memory stage. When a relation stages up or down in the memory hierarchy, these preprocessing mechanisms can function to select tuples and attributes needed by a query and to arrange the relation for parallel processing. SPIRIT-III provides three basic preprocessing filters, called as a whole the Tuple Stream Filter: the tuple selector, the attribute selector, and the grouping filter, implemented with a hash function, which rearranges an original relation and groups the relation into subrelations. The operation of this grouping filter is the primitive preprocessing operation for executing Join and Projection. Then, without the overhead of interprocessor communications, each microprocessor can execute relational algebra operations to a few subsegments assigned to it in parallel. Therefore, SPIRIT-III can perform Join and Projection operations by O (N/L) (L = number of microprocessors), whereas the early RDBMs required O (N X N/L). The proposed SPIRIT-III, which includes features from data-staging architecture to relational algebra execution architecture under the total concept, is the most powerful RDBM based on the state of the art.