On parallel retrievals using distributed memory computers

Abstract

This paper introduces an. eficient indexing organization. which provides for parallel retrievals. G,iven. a p-processor massively parallel computer! our orymizzation allouts a m a x i m u m of p retrievals to be performed con.cu.rrently and thzls achieves an. approz*imate speed ufp o f p in, the best case. T h e index structure i s based o n the multi-B-tree index stru.cture. It .is compatible ,with cu.rrent automatic retrieval and qu~ery process-.ing techniques. I t is also compatible with a variety of file organizations. In add.ition to a n indexed attribute which w e refer t o as primary attribute, a secondary attribute can be included in, the index at n o additional cost. Updates are achie.ved with a high degree of e$;-c i m c y as well. T h e parallel organ,ization of a multi-B-tree index improves conuiderablg the performan.ce of the index. T h e method discussed here can be easily applied t o standard B+ tree, B* tree and other B tree variations.