Capacity Planning for Web Performance: Metrics, Models, and Methods - Softcover

Almeida, Virgillio; Almeida, Virgilio A. F.

Softcover

ISBN 10: 0136938221 ISBN 13: 9780136938224

Publisher: Prentice Hall, 1998

Synopsis

Uses quantitative methods to analyze client/server and Web-based applications, leading the capacity planner in a step-by-step fashion through the process of determining the most cost-effective system configurations and networking architectures. CD-ROM included. Paper. DLC: Computer capacity - PLanning.

About this title

About the Author

DANIEL A. MENASCE is a Professor of Computer Science at George Mason University, VA. He has published extensively in the area of performance modeling, client/system performance evaluation, and software performance engineering. Menasce was elected a Fellow of the Association for Computing Machinery (ACM) in recognition of outstanding contributions to information technology.

VIRGILIO A. F. ALMEIDA is a Professor of Computer Science at the Federal University of Minas Gerais (UFMG), Brazil. He has published extensively in the area of distributed systems and World Wide Web performance. Almeida held visiting faculty and research positions at Boston University and XEROX PARC.

From the Back Cover

"This excellent book presents a new way to model, analyze, and plan for these new performance problems associated with the Web's bursty and highly-skewed load characteristics. A valuable resource for students and for Web administrators." -- Jim Gray, Senior Researcher, Microsoft Research

"Many have said that the Web is too amorphous and chaotic to permit meaningful performance forecasts. Menasce and Almeida demolish this myth. Throughput, response time, and congestion can be measured and predicted, all using familiar tools from queueing networks that you can run on your own computer. There is no other book like this. It is a first." -- Peter J. Denning, Professor of Computer Science, George Mason University and former President of the ACM

"This book takes the mystery out of analyzing Web performance. The authors have skillfully culled through more than a 25 years of research, and selected the results most critical to Web performance, and developed important new material that deals directly with the special properties of applications that run on the Web. With everything together in a single volume, Menasce and Almeida have created a superb starting point for anyone wishing to explore the world of Web performance." -- Jeffrey P. Buzen, Chief Scientist and CoFounder, BGS Systems

"This is a welcome approach to the performance analysis of today's web-based Internet. It is a useful and practical treatment that is eminently accessible to the non-mathematical professional. An impressive feature the authors provide is to deal directly with the fractal nature of web-based traffic; no simple and practical treatment has been offered before, and theirs is a timely contribution." -- Leonard Kleinrock, Professor of Computer Science, UCLA

As more and more businesses rely on distributed client/server and Web-based applications, performance considerations become extremely important. Capacity Planning for Web Performance uses quantitative methods to analyze these systems. It leads the capacity planner, in a step-by-step fashion, through the process of determining the most cost-effective system configurations and networking architectures. The quantitative methods lead to the development of performance-predictive models for capacity planning. Instead of relying on intuition, ad hoc procedures, and rules of thumb, Capacity Planning for Web Performance provides a uniform and sound way for dealing with performance problems. A large number of numeric and practical examples help the reader understand the quantitative approach adopted here.

Includes a CD-ROM containing several Microsoft Excel(r) workbooks supported by Visual Basic(r) modules, samples of http logs, and programs to process them. The Excel workbooks allow the readers to immediately put into practice the methods and models discussed here.

Includes the following tools for analyzing client/server systems, intranets, and Internet Web sites:

* Performance-oriented analysis of network protocols
* Modeling of delays
* Workload characterization and forecasting
* Use of industry-standard benchmarks
* Queuing network-based models

From the Inside Flap

Preface

The client/server model emerged out of the convergence of computers and communications, the availability of inexpensive and powerful desktop computers with Graphical User Interfaces (GUI) and multimedia presentation devices, and advanced forms of data input, including voice input. By bringing the computation closer to the user and his/her business processes and by providing richer and easier to use interfaces, client/server systems can increase productivity, increase customer satisfaction, and cut down significantly on training costs.

A special and important case of client/server applications is the World Wide Web (WWW) which has been growing at an enormous pace. The number of Web sites doubles every six months. Applications such as digital libraries, electronic commerce, video on-demand, and distance learning increase Internet and Web traffic at even higher rates. Virtual stores on the Web allow you to buy cars, books, computers, and many other products and services.

As more and more businesses and government agencies rely on distributed client/server and Web-based applications, performance considerations become extremely important. Performance analysis of intranets and Web servers is unique in many senses. First, the number of WWW clients is in the tens of millions and growing. The randomness associated with the way users visit pages and request Web services makes the problem of workload forecasting and capacity planning difficult.

This book uses a quantitative approach to analyzing client/server and Web-based systems. This approach lends itself to the development of performance predictive models for capacity planning. Instead of relying on intuition, ad hoc procedures, and rules of thumb, we provide a uniform and formal way for dealing with performance problems. The performance models discussed here are based on the theory of queuing networks. Another important feature of the book is its treatment of two fundamental characteristics of WWW traffic: burstiness and heavy-tailed distributions of Web document sizes. The techniques and models presented here provide a simple and practical approach to dealing with this type of environment.

Although some of the concepts about client/server and Web architectures may be familiar to some of the readers, these concepts are revisited here in light of quantitative and performance issues. A large number of numeric and practical examples help the reader understand the quantitative approach adopted in this book. Several MS Excel workbooks supported by Visual Basic modules accompany the book. These workbooks allow the readers to immediately put into practice the methods and models discussed here.

Who should read this book

Information technology managers and related staff must guarantee that the networked services under their management provide an acceptable quality of service to their users. Managers must avoid the pitfalls of inadequate capacity and meet users' performance expectations in a cost-effective manner. System administrators, Web masters, network administrators, capacity planners and analysts, IT managers and consultants, will benefit from reading parts or the entire book. Its practical, yet sound and formal, approach provides the grounds for understanding modern and complex networked environments.

This book can be used as a textbook for senior undergraduate and graduate courses in Computer Science and Computer Engineering. At the undergraduate level, the book is a good starting point to motivate students to learn the important implications and solutions to performance problems.

At the graduate level, it can be used in conjunction with the book {\em Capacity Planning and Performance Modeling: From Mainframes to Client-Server Systems\/}, by Menasc\'e, Almeida, and Dowdy, Prentice Hall, 1994, in System Performance Evaluation courses. The combination of these two books offers a theoretical and practical foundation in performance modeling. The present book can also be used as a supplement for systems courses, including Operating Systems, Distributed Systems, and Networking, both at the undergraduate and graduate levels.

Book Organization

{\em Chapter 1\/} introduces, through a series of examples, the importance of

performance considerations in client/server and Web-based systems.

The concepts of capacity planning, service levels, and workload evolution are defined and illustrated.

{\em Chapter 2\/} presents a brief discussion on local and wide area networks and their protocols including TCP/IP, Ethernet, Token Ring, and FDDI. The chapter revisits the client server (C/S) computing paradigm and discusses several kind of servers, such as file servers, database servers, application servers, groupware servers, object servers, and Web servers. Various architectural client/server issues including fat versus thin clients, two-tier versus three-tier C/S architectures are presented. These concepts are presented in light of performance considerations and tradeoffs.

{\em Chapter 3\/} investigates, in detail, the nature of the delay incurred by a typical C/S transaction through the use of communication-processing delay diagrams. This chapter shows how service times can be computed at single disks, disk arrays, networks, and routers. Queues and contention are defined more formally and some very basic and important performance results from Operational Analysis are introduced.

{\em Chapter 4\/} discusses issues that affect performance of Web servers and intranets. The chapter starts by looking at the sources of delay in Web environments. After discussing the end-user perspective to the Internet and intranet performance, the chapter provides an assessment. It examines the components and protocols involved in the execution of Web services and analyzes their capacity and performance issues.

{\em Chapter 5\/} introduces a step-by-step methodology to determine the most cost-effective system configuration and networking architecture. The main steps of the methodology are: understanding the environment, workload characterization, workload model validation and calibration, performance model development, performance model validation and calibration, workload forecasting, performance prediction, cost model development, cost prediction, and cost/performance analysis.

{\em Chapter 6\/} describes and illustrates with examples the major steps required for the construction of workload models. The common steps to be followed by any workload characterization project include specification of a point of view from which the workload will be analyzed, choice of the set of parameters that capture the most relevant characteristics of the workload for the purpose of the study, monitoring the system to obtain the raw performance data, analysis and reduction of performance data, and construction of a workload model.

{\em Chapter 7\/} presents several different standard industry benchmarks---such as TPC-C and TPC-D, SPECcpu, SPECweb, LADDIS, and WebStone---and shows how to use benchmark results as a complementary source of information to support the capacity planning methodology.

{\em Chapter 8\/} starts by introducing very simple models of network-based systems. Complexity is progressively introduced and the solution to each model is presented using first principles and intuitive concepts. After a few models are presented, the approach is generalized. The models presented in this chapter are called system-level performance models since they view the system being modeled as a ``black box''.

{\em Chapter 9\/} introduces powerful techniques to analyze the performance of intranets and client/server systems. The chapter considers the components that make up a networked system. Component-level models account for the different resources of the system and the way they are used by different requests. The solution techniques for component-level models are based on Queuing Networks (QNs). Solution methods for both open and closed QNs with multiple classes of customers are presented in this chapter.

(\em Chapter 10\/} shows how the performance models discussed in previous chapters can be specialized to handle Web performance issues. The chapter also shows how certain special features common in Web workloads, such as burstiness and heavy tail distributions of file sizes, can be accounted for in performance models for the Web.

{\em Chapter 11\/} discusses aspects and techniques for workload forecasting, including regression models, moving averages, and exponential smoothing. The steps of the workload forecasting process discussed here include selection of the workload to be forecast, analysis of historical data and estimation of workload growth, selection of a forecasting technique, application of the forecasting technique to the historical data, and analysis and validation of forecast results.

{\em Chapter 12\/} presents a framework for collecting performance data in networked environments that use the C/S computing paradigm. Next, it discusses the main issues in the process of measuring the performance of network-based systems. The chapter does not focus on any specific product or manufacturer. Instead, it presents general procedures for transforming typical measurement data into input parameters. The procedures can be thought of as a set of major guidelines for obtaining input parameters for performance models.

{\em Chapter 13\/} wraps up the book by emphasizing the approach and methods used to develop and solve analytical performance models of client/server and Web-based systems. It also provides a unified view of the entire book.

{\em Appendix A\/} contains a glossary of the important terms introduced in the book.

{\em Appendix B\/} provides a quick reference manual of the MS Excel workbooks that accompany this book and describes the C program provided to derive burstiness parameters for HTTP logs.

\subsubsection{Acknowledgments}

The authors would like to thank their many colleagues for discussions that contributed substantially to this book.

Special thanks go to our co-author in the first book of this series, Larry Dowdy of Vanderbilt University, for his enthusiasm and dedication to the field. Particular thanks go to Jeff Buzen of BGS Systems, Peter Denning of George Mason University, Jim Gray of Microsoft Research, and Leonard Kleinrock of UCLA for their praise of our work and permission to quote them on the cover. Jim Gray's suggestions on the manuscript are deeply appreciated. We would also like to thank our Acquisitions Editor at Prentice Hall PTR, Stephen Solomon, his assistant, Bart Blanken, our Production Editor, Craig Little, our Copyeditor, Kathy Finch, and our Marketing Manager, Dan Rush, for their support during the preparation and production of this book. Daniel Menasce would like to thank his students and colleagues at George Mason University for providing a stimulating work environment. He would also like to thank his parents for their love and guidance in life. Special recognition goes to his wife Gilda for a life full of love and companionship and to his children Flavio and Juliana for being a constant source of joy. Virgilio Almeida would like to thank his colleagues and students at UFMG and Boston University. In particular, he wants to thank Wagner Meira, Andre Cinelli, Cristina Murta, Jussara Almeida, Mark Crovella, Azer Bestravos, David Yates, and Carlos Cunha. Finally, Virgilio would also like to express his gratitude to his family, parents (in memoriam), brothers, and many relatives and friends. His wife Rejane and sons Pedro and Andre have always been a source of love, sweetness, joy, emotional support, and continuous encouragement.

\clearpage \subsubsection{Book's Web site and authors' addresses}

A Web site will be maintained at cs.gmu/$\tilde{~}$menasce/webbook/ to keep the readers informed about new developments related to the book.
The authors' e-mail and postal addresses and Web sites are:
\makebox{}\\ \noindent Prof. Daniel A. Menasc\'e\\ Department of Computer Science, MS 4A5\\ George Mason University\\ Fairfax, VA 22030-4444\\ (703) 993-1537\\ menasce@cs.gmu\\ cs.gmu/faculty/menasce.html \noindent\\
Prof. Virgilio A. F. Almeida\\ Department of Computer Science \\ Universidade Federal de Minas Gerais\\ P.O. Box 920\\ 31270-010 Belo Horizonte, MG\\ Brazil\\ +55 31 499-5887\\ virgilio@dcc.ufmg.br\\ dcc.ufmg.br/$\tilde{~}$virgilio

Edition details

• Publisher: Prentice Hall
• Publication date: 1998
• Language: English
• ISBN 10: 0136938221
• ISBN 13: 9780136938224
• Binding: Paperback
• Number of pages: 350