The impact of seeking in partial match retrieval

Abstract

We atend the problem of partial match query satisfaction by studying the impact of seeking. We study the physical location of the autput pages by considenng the number and the sparseness of cylinders holding the resulting pages. Lawer and upper seek time bunds, as well as the average behavior of the seek time are adculuted by assuming m m e real fi9ures of specific modern disk system devices. m e main conclusion is that the seek time is a factor aflecting the partial match query .response time and needs to be included in the overall performance measuring. roduct ionA multi-attribute file is a collection of records characterized by more than one attribute, divided into pages and then stored in one or several disks. A partial match query (pm query) to a multi-attribute file is a request specifying certain values for one or several attributes. In such a case all pages containing at least one record having attribute values matching with the d u e s specified by the pm query must be accessed. The problem of partial match retrieval on a multi-attribute file, residing on a single disk. has been examined in order to design an efficient allocation scheme minimizing the required number of disk accesses. More recently, a similar problem has been tackled by considering a set of parallel disk instead of a single one. The multi-disk file problem aims at grouping file records in pages and allocating pages in parallel disks so that a pm query is efficiently answered by maximizing concurrence of disk accesses. Several data allocation methods have been studied lately in order to deal with the problem of assigning pages to one or several disks. The most frequent one is the Disk Modulo (DM) allocation method. which has been proven to be more effective because of its simplicity and its "strictly optimality" property (guaranteed under conditions). Variations of the DM method have been examined in [4,6] and optimal behavior is investigated under certain assumptions. In [3,7] Gray Codes are used in order to distribute binary Cartesian product files on multiple disks, while in [SI Error Correcting Codes have been introduced for file allocation. In all the previous works, the performance i s measured in terms d the maximum number of pages that have to be accessed concurrently. Exceptionally, in [7] the number of clusters (i.e. a set of qualifying pages which are physically stored consecutively) is taken as a performance indicator. Summarizing, until now either the number of pages or the number of clusters have been taken as the crucial measure influencing performance's estimation.In the present work we consider the seek time needed to respond to a query since any retrieval operation includes the time required to move to particular cylinders holding the required pages. " h s , the Partial Match Iietrieval problem is studied under the new perspective of not having the pages consecutively stored on the disk, ie. they are stored at different disk cylinders. Here, we assume that a SCAN (or CSCAN) disk scheduling policy is used, according to wh...