THE FRACTAL STRUCTURE OF DATA REFERENCE- P26:For purposes of understanding its performance, a computer system is traditionally viewed as a processor coupled to one or more disk storage devices, and driven by externally generated requests (typically called transactions). Over the past several decades, very powerful techniques have become available to the performance analyst attempting to understand, at a high level, the operational behavior of such systems. | Disk Applications A Statistical View 115 The initial two sections of the chapter introduce the basic structure of the deployable applications model and examine the calculation of application characteristics. The final two sections then turn to the implications ofthe model with respect to disk performance requirements seeking a common ground between the two contrasting views outlined at the beginning ofthe chapter. 1. DEPLOYABLE APPLICATIONS MODEL Consider an application a with the following requirements va transaction volume transactions per second . Sa storage gigabytes . The purpose of the deployable applications model is to estimate whether such an application will be worthwhile to deploy at any given time. Thus we must consider both the benefit of deploying the application as well as its costs. Application a will be considered deployable if its costs are no larger than its estimatedbenefits. The benefit ofapplicationa is tied to desired events in the real world such as queries being answered purchases being approved or orders being taken. Such real-world events typically correspond to database transactions. Therefore we estimate the dollar benefit ofapplication a from its transaction volume benefit. 9. where k is a constant. For the sake ofsimplicity we divide application costs intojusttwo categories. Transaction processing costs including cpu costs and hardware such as point-of-sale terminals or network bandwidth upgrades are accounted for based upon transactionvolume - . where k2 kj. To account for the storage costs of application a we examine the resources needed to meet both its storage and i o requirements. Its storage requirements have already been stated as equal to sa. In keeping with the transaction-based scheme of and we characterize application a si o requirement in i o s per second as Gv where G is a constant for simple transactions G tends to be in the area of 10-20 I O s per transaction . Against the requirements of the application as
