Catálogo de publicaciones - libros

Time is the basis of all computer performance management (Fig. 1.1). It is so fundamental that it could be called the zeroth-order performance metric. In the context of computer performance analysis, time manifests itself in a multitude of metrics like service time, response time, round-trip time, memory latency, and mean time to failure, to name just a few. In view of this variety, it would seem fitting to open a book on performance analysis with a review of these various performance metrics. Surprisingly, there seems to be no precedent for such a discussion in any of the readily available textbooks on computer performance analysis. It is about time someone provided a brief discourse about time, and that is what we offer in this chapter.

Think about the times you have had to wait in line because other people wanted the same thing you did, and you get that sinking feeling so often associated with the phenomenon of queueing, the bane of modernity. You queue while commuting to work, while boarding an aircraft, at a grocery store, at the post office, the doctor’s office or connecting to a Web site (TCP listen queue).

In Chap. 2 the analysis of queueing performance only involved a single queueing center. Even when more than one queueing center was available, the customer only visited one of them. Such single queueing centers can only be used successfully to represent a single device or elements of a computer system, e.g., a disk device. If the performance analyst is required to assess the interaction of various devices in the complete computer system, the analysis methods presented in Chap. 2 are not sufficient, in general.

The term means different things to different people. For example, it might imply the number of to a system administrator, whereas we saw in Chap. 2 that it tends to imply to a performance analyst.

The material presented here comes from real case studies. Only the names, place, and numbers have been changed to protect the guilty. In this chapter you will see the power of bounds analysis based on the underlying queueing concepts presented in Chaps. 2 and 3. The main point is to demonstrate how much can be determined about performance without knowing the detailed performance characteristics.

This chapter introduces the PDQ© (Pretty Damn Quick) queueing analyzer and explains how to use it. We begin with some guidelines on how to build performance models and then move on to the specifics of the PDQ library in Perl. Finally, we present the actual PDQ codes for the examples discussed in Chaps. 2 and 3. Other PDQ model codes are embedded in the chapters of Part II. Instructions for installing PDQ and creating the corresponding Perl module can be found in Appendix F.

In this chapter we turn to the subclass of multicomputer architectures known as (SMP). Because of their intrinsic economy, expandability, performance, and reliability, SMPs have found their way into a wide range of applications. In particular, commercial applications, and that is the focus of this chapter. Distinct from scientific requirements, the emphasis in the commercial arena is on high levels of coarse-grain concurrency rather than fine-grain parallelism.

There is a story about three blind men and an elephant that goes like this.

In this chapter we present performance analysis for client/server architectures using PDQ. This material is the most complex use of PDQ so far in that it draws on the techniques presented in previous chapters and extends them to software and communication network analysis.

In this chapter we examine the performance characteristics of the latest innovation in client/server technology—Web technology. Unlike the traditional client/server systems discussed in Chap. 9, each Web client typically makes high-frequency, short-term accesses to a relatively small number of servers.