A practitioner's guide to addressing algorithms

Abstract

This paper consolidates a number of popular rules of thumb which have been suggested for the design of record addressing algorithms, and discusses the applicability of these rules to large commercial databases. Guidelines for selecting identifier transformations, overflow techniques, loading factors, bucket sizes, and loading order are considered. Particular attention is focused on the reasonableness of common heuristics for determining primary or secondary bucket sizes. A mathematical model which explicitly considers storage device characteristics and time/space cost tradeoffs is used to analyze the effect of design parameters on overall system costs. A specific design example is presented and solved.

Optimal file designs and reorganization points

“…Also, let NcF' be the number of records in a file whose descriptor is F. To characterize the usage of a file, we introduce the following parameters which are used to specify the expected execution frequencies [8], some execution frequency statistics (e.g., fret and fupd) are expressed in units of "per record per tune period," rather than just "per time period." Except to be consistent with previous models, there does not seem to be any a priori reason why one unit of frequency is better than another.…”

“…However, more general file design problems can be (and have been) addressed by allowing other parameters of a fde descriptor, specifically the design parameters of Table I, to be optimization variables. For example, optimization variables in studies by Severance and Duhne [8] and Van der Pool [12] were the block capacities R and Ro and the initial loading factor s. Optimization involving the file structure parameter Fs was recently considered by Batory [3]. The scope of the optimization could be further enlarged by including other structures besides indexed-sequential and hash-based files.…”

“…Efficient designs for these files have been studied in two different environments: load time and steady state. The load time environment of Severance and Duhne [8] was retrieval only; there were no record updates, insertions, or deletions. The steadystate environment of Van der Pool [12] was also retrieval only, where a large, but equal, number of insertions and deletions had occurred prior to retrieval.…”

“…The steadystate environment of Van der Pool [12] was also retrieval only, where a large, but equal, number of insertions and deletions had occurred prior to retrieval. Performance deterioration was shown to be minimal in [12], while deterioration did not occur in [8]. As a consequence, it was unnecessary to consider file reorganizations.…”

Differential files

“…If the two data files and their search mechanisms can be assigned to different devices and accessed via separate channels , then both file searche s may proceed in parallel and system users wil l not perceive the increased delay. Wisen Such overlap is i mpossible , one can expect the average time of a data record retrieval (assumi ng a Judicious selection of the di fferential fxl e search strategy [15)) to increase by the amount of time reguired for a random access to secondary memory.…”