A Framework for Distributing and Parallelizing Secondary Mem8ry Search Structures a Framework for Distributing and Parallelizing Secondary Mem8ry Search Structures a Framework for Distributing and Parallelizlng Secondary

Introduction Traditional architectures which consist of a single processor, a small main memory, and secondary memory devices (e.g. disks), are gradually being replaced by multiprocessor multi-disk architec­ tures with a relatively large main m~mory. The processors either reside in the same site and communicate via the shared memory, or reside in different sites and communicate via a local network. We consider architectures which consist of a relatively small number of processors P (say 2-20), and each processor possesses a disk (or a group of disks). A fundamental issue when designing file systems in such architectures is the design of search structures for main taining single key files, multi-key files, and indices. The search structures may be distributed among the processor-disk pairs in order to achieve parallel execution capabilities, and an effort should be made to reduce the consumption of resources such as main memory. The design and implementation of traditional search structures are well understood. Exam­ ples of such traditional search structures are B+-trees and their variants ~C], dense indices [U], Extendible Hash tables (FNPS], Bounded Disordered Files [L02], TT-trees [L], k-d trees [B], Grid Files [:N"HS], K-D-B-trees [R], hB-trees [LS] etc. Therefore, the standard approach is to keep using the traditional search structures within the modern architecture, while making an effort towards a distribution of the structure's components among the processors and disks, in order to achieve parallel execution capabilities [LB]. A similar approach [DGGHKM] [MS1] is to equally partition the file records among processors (for example, by usmg a hash function), each of which maintains its part in a local traditional search structure, stored in its disk. We argue that addi­ tional benefits (besides parallel capabilities) can be achieved by modifying the tradi tional search structures, in order to achieve a better fit to multiprocessor multi-disk environments. In this paper, we propose a framework for distributing search structures in multiprocessor multi-disk environments. Given a traditional search structure, its distribution process, according to our framework, consists of three parts: 1. Structural design: The components of the search structure l are modified by changing their structure (various components may be modified differently). 2. Mapping function: A storage device is determined for each of the search structure's components. An example of a mapping of a component is as follows: 1 An example of a search structure component is a node in a search tree. 1 1 1 An example of a …