Visit https://ebookultra.

com to download the full version and

explore more ebooks or textbooks

Performance modeling and design of computer

systems queueing theory in action Harchol-Balter

_____ Click the link below to download _____

https://ebookultra.com/download/performance-modeling-and-
design-of-computer-systems-queueing-theory-in-action-
harchol-balter/

Explore and download more ebooks or textbooks at ebookultra.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Queueing Networks and Markov Chains Modeling and

Performance Evaluation With Computer Science Applications
Second Edition Gunter Bolch
https://ebookultra.com/download/queueing-networks-and-markov-chains-
modeling-and-performance-evaluation-with-computer-science-
applications-second-edition-gunter-bolch/

Human Computer Interaction and Operators Performance

Optimizing Work Design with Activity Theory Ergonomics
Design and Management Theory and Applications 1st Edition
Gregory Z. Bedny
https://ebookultra.com/download/human-computer-interaction-and-
operators-performance-optimizing-work-design-with-activity-theory-
ergonomics-design-and-management-theory-and-applications-1st-edition-
gregory-z-bedny/

Advances in Modeling and Design of Adhesively Bonded

Systems 1st Edition S. Kumar

https://ebookultra.com/download/advances-in-modeling-and-design-of-
adhesively-bonded-systems-1st-edition-s-kumar/

Design and modeling for computer experiments 1st Edition

Kai-Tai Fang

https://ebookultra.com/download/design-and-modeling-for-computer-
experiments-1st-edition-kai-tai-fang/
Fundamentals of Performance Evaluation of Computer and
Telecommunications Systems 1st Edition Mohammed S. Obaidat

https://ebookultra.com/download/fundamentals-of-performance-
evaluation-of-computer-and-telecommunications-systems-1st-edition-
mohammed-s-obaidat/

Informed systems organizational design for learning in

action 1st Edition Somerville

https://ebookultra.com/download/informed-systems-organizational-
design-for-learning-in-action-1st-edition-somerville/

Computer Design and Computational Defense Systems 1st

Edition Nikos E. Mastorakis

https://ebookultra.com/download/computer-design-and-computational-
defense-systems-1st-edition-nikos-e-mastorakis/

Theory of Quantum and Classical Connections In Modeling

Atomic Molecular And Electrodynamical Systems 1st Edition
Alexandru Popa
https://ebookultra.com/download/theory-of-quantum-and-classical-
connections-in-modeling-atomic-molecular-and-electrodynamical-
systems-1st-edition-alexandru-popa/

Design for Reliability Information and Computer Based

Systems 1st Edition Eric Bauer

https://ebookultra.com/download/design-for-reliability-information-
and-computer-based-systems-1st-edition-eric-bauer/
Performance modeling and design of computer systems
queueing theory in action Harchol-Balter Digital Instant
Download
Author(s): Harchol-Balter, Mor
ISBN(s): 9781107027503, 1107027500
Edition: Rep
File Details: PDF, 8.26 MB
Year: 2014
Language: english
more information - www.cambridge.org/9781107027503
 Performance Modeling and Design of Computer Systems
Computer systems design is full of conundrums:
r Given a choice between a single machine with speed s, or n machines each with
speed s/n, which should we choose?
r If both the arrival rate and service rate double, will the mean response time stay the
same?
r Should systems really aim to balance load, or is this a convenient myth?
r If a scheduling policy favors one set of jobs, does it necessarily hurt some other jobs,
or are these “conservation laws” being misinterpreted?
r Do greedy, shortest-delay, routing strategies make sense in a server farm, or is what
is good for the individual disastrous for the system as a whole?
r How do high job size variability and heavy-tailed workloads affect the choice of a
scheduling policy?
r How should one trade off energy and delay in designing a computer system?
r If 12 servers are needed to meet delay guarantees when the arrival rate is 9 jobs/sec,
will we need 12,000 servers when the arrival rate is 9,000 jobs/sec?
Tackling the questions that systems designers care about, this book brings queueing theory
decisively back to computer science. The book is written with computer scientists and
engineers in mind and is full of examples from computer systems, as well as manufacturing
and operations research. Fun and readable, the book is highly approachable, even for
undergraduates, while still being thoroughly rigorous and also covering a much wider span
of topics than many queueing books.
Readers benefit from a lively mix of motivation and intuition, with illustrations, examples,
and more than 300 exercises – all while acquiring the skills needed to model, analyze,
and design large-scale systems with good performance and low cost. The exercises are an
important feature, teaching research-level counterintuitive lessons in the design of computer
systems. The goal is to train readers not only to customize existing analyses but also to
invent their own.

Mor Harchol-Balter is an Associate Professor in the Computer Science Department at

Carnegie Mellon University. She is a leader in the ACM Sigmetrics Conference on Measure-
ment and Modeling of Computer Systems, having served as technical program committee
chair in 2007 and conference chair in 2013.
Performance Modeling and
Design of Computer Systems

Queueing Theory in Action

Mor Harchol-Balter
Carnegie Mellon University, Pennsylvania
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Mexico City
Cambridge University Press
32 Avenue of the Americas, New York, NY 10013-2473, USA
www.cambridge.org
Information on this title: www.cambridge.org/9781107027503


C Mor Harchol-Balter 2013

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 2013

Printed in the United States of America

A catalog record for this publication is available from the British Library.

Library of Congress Cataloging in Publication Data

Harchol-Balter, Mor, 1966–
Performance modeling and design of computer systems : queueing theory in
action / Mor Harchol-Balter.
pages cm
Includes bibliographical references and index.
ISBN 978-1-107-02750-3
1. Transaction systems (Computer systems) – Mathematical models. 2. Computer
systems – Design and construction – Mathematics. 3. Queueing theory.
4. Queueing networks (Data transmission) I. Title.
QA76.545.H37 2013
519.8 2–dc23 2012019844

ISBN 978-1-107-02750-3 Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of

URLs for external or third-party Internet websites referred to in this publication and
does not guarantee that any content on such websites is, or will remain, accurate or
appropriate.
To my loving husband Andrew, my awesome son Danny,
and my parents, Irit and Micha
I have always been interested in finding better designs for computer systems, designs
that improve performance without the purchase of additional resources. When I look
back at the problems that I have solved and I look ahead to the problems I hope to
solve, I realize that the problem formulations keep getting simpler and simpler, and my
footing less secure. Every wisdom that I once believed, I have now come to question:
If a scheduling policy helps one set of jobs, does it necessarily hurt some other jobs,
or are these “conservation laws” being misinterpreted? Do greedy routing strategies
make sense in server farms, or is what is good for the individual actually disastrous for
the system as a whole? When comparing a single fast machine with n slow machines,
each of 1/nth the speed, the single fast machine is typically much more expensive – but
does that mean that it is necessarily better? Should distributed systems really aim to
balance load, or is this a convenient myth? Cycle stealing, where machines can help
each other when they are idle, sounds like a great idea, but can we quantify the actual
benefit? How much is the performance of scheduling policies affected by variability
in the arrival rate and service rate and by fluctuations in the load, and what can we do
to combat variability? Inherent in these questions is the impact of real user behaviors
and real-world workloads with heavy-tailed, highly variable service demands, as
well as correlated arrival processes. Also intertwined in my work are the tensions
between theoretical analysis and the realities of implementation, each motivating the
other. In my search to discover new research techniques that allow me to answer
these and other questions, I find that I am converging toward the fundamental core
that defines all these problems, and that makes the counterintuitive more believable.
 Contents

Preface xvii
Acknowledgments xxiii

I Introduction to Queueing
1 Motivating Examples of the Power of Analytical Modeling 3
1.1 What Is Queueing Theory? 3
1.2 Examples of the Power of Queueing Theory 5
2 Queueing Theory Terminology 13
2.1 Where We Are Heading 13
2.2 The Single-Server Network 13
2.3 Classification of Queueing Networks 16
2.4 Open Networks 16
2.5 More Metrics: Throughput and Utilization 17
2.6 Closed Networks 20
2.6.1 Interactive (Terminal-Driven) Systems 21
2.6.2 Batch Systems 22
2.6.3 Throughput in a Closed System 23
2.7 Differences between Closed and Open Networks 24
2.7.1 A Question on Modeling 25
2.8 Related Readings 25
2.9 Exercises 26

II Necessary Probability Background

3 Probability Review 31
3.1 Sample Space and Events 31
3.2 Probability Defined on Events 32
3.3 Conditional Probabilities on Events 33
3.4 Independent Events and Conditionally Independent Events 34
3.5 Law of Total Probability 35
3.6 Bayes Law 36
3.7 Discrete versus Continuous Random Variables 37
3.8 Probabilities and Densities 38
3.8.1 Discrete: Probability Mass Function 38
3.8.2 Continuous: Probability Density Function 41
3.9 Expectation and Variance 44
3.10 Joint Probabilities and Independence 47

vii
viii contents

3.11 Conditional Probabilities and Expectations 49

3.12 Probabilities and Expectations via Conditioning 53
3.13 Linearity of Expectation 54
3.14 Normal Distribution 57
3.14.1 Linear Transformation Property 58
3.14.2 Central Limit Theorem 61
3.15 Sum of a Random Number of Random Variables 62
3.16 Exercises 64
4 Generating Random Variables for Simulation 70
4.1 Inverse-Transform Method 70
4.1.1 The Continuous Case 70
4.1.2 The Discrete Case 72
4.2 Accept-Reject Method 72
4.2.1 Discrete Case 73
4.2.2 Continuous Case 75
4.2.3 Some Harder Problems 77
4.3 Readings 78
4.4 Exercises 78
5 Sample Paths, Convergence, and Averages 79
5.1 Convergence 79
5.2 Strong and Weak Laws of Large Numbers 83
5.3 Time Average versus Ensemble Average 84
5.3.1 Motivation 85
5.3.2 Definition 86
5.3.3 Interpretation 86
5.3.4 Equivalence 88
5.3.5 Simulation 90
5.3.6 Average Time in System 90
5.4 Related Readings 91
5.5 Exercise 91

III The Predictive Power of Simple Operational Laws: “What-If”

Questions and Answers
6 Little’s Law and Other Operational Laws 95
6.1 Little’s Law for Open Systems 95
6.2 Intuitions 96
6.3 Little’s Law for Closed Systems 96
6.4 Proof of Little’s Law for Open Systems 97
6.4.1 Statement via Time Averages 97
6.4.2 Proof 98
6.4.3 Corollaries 100
6.5 Proof of Little’s Law for Closed Systems 101
6.5.1 Statement via Time Averages 101
6.5.2 Proof 102
6.6 Generalized Little’s Law 102
 contents ix

6.7 Examples Applying Little’s Law 103

6.8 More Operational Laws: The Forced Flow Law 106
6.9 Combining Operational Laws 107
6.10 Device Demands 110
6.11 Readings and Further Topics Related to Little’s Law 111
6.12 Exercises 111
7 Modification Analysis: “What-If” for Closed Systems 114
7.1 Review 114
7.2 Asymptotic Bounds for Closed Systems 115
7.3 Modification Analysis for Closed Systems 118
7.4 More Modification Analysis Examples 119
7.5 Comparison of Closed and Open Networks 122
7.6 Readings 122
7.7 Exercises 122

IV From Markov Chains to Simple Queues

8 Discrete-Time Markov Chains 129
8.1 Discrete-Time versus Continuous-Time Markov Chains 130
8.2 Definition of a DTMC 130
8.3 Examples of Finite-State DTMCs 131
8.3.1 Repair Facility Problem 131
8.3.2 Umbrella Problem 132
8.3.3 Program Analysis Problem 132
8.4 Powers of P: n-Step Transition Probabilities 133
8.5 Stationary Equations 135
8.6 The Stationary Distribution Equals the Limiting Distribution 136
8.7 Examples of Solving Stationary Equations 138
8.7.1 Repair Facility Problem with Cost 138
8.7.2 Umbrella Problem 139
8.8 Infinite-State DTMCs 139
8.9 Infinite-State Stationarity Result 140
8.10 Solving Stationary Equations in Infinite-State DTMCs 142
8.11 Exercises 145
9 Ergodicity Theory 148
9.1 Ergodicity Questions 148
9.2 Finite-State DTMCs 149
9.2.1 Existence of the Limiting Distribution 149
9.2.2 Mean Time between Visits to a State 153
9.2.3 Time Averages 155
9.3 Infinite-State Markov Chains 155
9.3.1 Recurrent versus Transient 156
9.3.2 Infinite Random Walk Example 160
9.3.3 Positive Recurrent versus Null Recurrent 162
9.4 Ergodic Theorem of Markov Chains 164
x contents

9.5 Time Averages 166

9.6 Limiting Probabilities Interpreted as Rates 168
9.7 Time-Reversibility Theorem 170
9.8 When Chains Are Periodic or Not Irreducible 171
9.8.1 Periodic Chains 171
9.8.2 Chains that Are Not Irreducible 177
9.9 Conclusion 177
9.10 Proof of Ergodic Theorem of Markov Chains∗ 178
9.11 Exercises 183
10 Real-World Examples: Google, Aloha, and Harder Chains∗ 190
10.1 Google’s PageRank Algorithm 190
10.1.1 Google’s DTMC Algorithm 190
10.1.2 Problems with Real Web Graphs 192
10.1.3 Google’s Solution to Dead Ends and Spider Traps 194
10.1.4 Evaluation of the PageRank Algorithm 195
10.1.5 Practical Implementation Considerations 195
10.2 Aloha Protocol Analysis 195
10.2.1 The Slotted Aloha Protocol 196
10.2.2 The Aloha Markov Chain 196
10.2.3 Properties of the Aloha Markov Chain 198
10.2.4 Improving the Aloha Protocol 199
10.3 Generating Functions for Harder Markov Chains 200
10.3.1 The z-Transform 201
10.3.2 Solving the Chain 201
10.4 Readings and Summary 203
10.5 Exercises 204
11 Exponential Distribution and the Poisson Process 206
11.1 Definition of the Exponential Distribution 206
11.2 Memoryless Property of the Exponential 207
11.3 Relating Exponential to Geometric via δ-Steps 209
11.4 More Properties of the Exponential 211
11.5 The Celebrated Poisson Process 213
11.6 Merging Independent Poisson Processes 218
11.7 Poisson Splitting 218
11.8 Uniformity 221
11.9 Exercises 222
12 Transition to Continuous-Time Markov Chains 225
12.1 Defining CTMCs 225
12.2 Solving CTMCs 229
12.3 Generalization and Interpretation 232
12.3.1 Interpreting the Balance Equations for the CTMC 234
12.3.2 Summary Theorem for CTMCs 234
12.4 Exercises 234
 contents xi

13 M/M/1 and PASTA 236

13.1 The M/M/1 Queue 236
13.2 Examples Using an M/M/1 Queue 239
13.3 PASTA 242
13.4 Further Reading 245
13.5 Exercises 245

V Server Farms and Networks: Multi-server, Multi-queue Systems

14 Server Farms: M/M/k and M/M/k/k 253
14.1 Time-Reversibility for CTMCs 253
14.2 M/M/k/k Loss System 255
14.3 M/M/k 258
14.4 Comparison of Three Server Organizations 263
14.5 Readings 264
14.6 Exercises 264
15 Capacity Provisioning for Server Farms 269
15.1 What Does Load Really Mean in an M/M/k? 269
15.2 The M/M/∞ 271
15.2.1 Analysis of the M/M/∞ 271
15.2.2 A First Cut at a Capacity Provisioning Rule for the M/M/k 272
15.3 Square-Root Staffing 274
15.4 Readings 276
15.5 Exercises 276
16 Time-Reversibility and Burke’s Theorem 282
16.1 More Examples of Finite-State CTMCs 282
16.1.1 Networks with Finite Buffer Space 282
16.1.2 Batch System with M/M/2 I/O 284
16.2 The Reverse Chain 285
16.3 Burke’s Theorem 288
16.4 An Alternative (Partial) Proof of Burke’s Theorem 290
16.5 Application: Tandem Servers 291
16.6 General Acyclic Networks with Probabilistic Routing 293
16.7 Readings 294
16.8 Exercises 294
17 Networks of Queues and Jackson Product Form 297
17.1 Jackson Network Definition 297
17.2 The Arrival Process into Each Server 298
17.3 Solving the Jackson Network 300
17.4 The Local Balance Approach 301
17.5 Readings 306
17.6 Exercises 306
18 Classed Network of Queues 311
18.1 Overview 311
18.2 Motivation for Classed Networks 311
xii contents

18.3 Notation and Modeling for Classed Jackson Networks 314

18.4 A Single-Server Classed Network 315
18.5 Product Form Theorems 317
18.6 Examples Using Classed Networks 322
18.6.1 Connection-Oriented ATM Network Example 322
18.6.2 Distribution of Job Classes Example 325
18.6.3 CPU-Bound and I/O-Bound Jobs Example 326
18.7 Readings 329
18.8 Exercises 329
19 Closed Networks of Queues 331
19.1 Motivation 331
19.2 Product-Form Solution 333
19.2.1 Local Balance Equations for Closed Networks 333
19.2.2 Example of Deriving Limiting Probabilities 335
19.3 Mean Value Analysis (MVA) 337
19.3.1 The Arrival Theorem 338
19.3.2 Iterative Derivation of Mean Response Time 340
19.3.3 An MVA Example 341
19.4 Readings 343
19.5 Exercises 343

VI Real-World Workloads: High Variability and Heavy Tails

20 Tales of Tails: A Case Study of Real-World Workloads 349
20.1 Grad School Tales . . . Process Migration 349
20.2 UNIX Process Lifetime Measurements 350
20.3 Properties of the Pareto Distribution 352
20.4 The Bounded Pareto Distribution 353
20.5 Heavy Tails 354
20.6 The Benefits of Active Process Migration 354
20.7 Pareto Distributions Are Everywhere 355
20.8 Exercises 357
21 Phase-Type Distributions and Matrix-Analytic Methods 359
21.1 Representing General Distributions by Exponentials 359
21.2 Markov Chain Modeling of PH Workloads 364
21.3 The Matrix-Analytic Method 366
21.4 Analysis of Time-Varying Load 367
21.4.1 High-Level Ideas 367
21.4.2 The Generator Matrix, Q 368
21.4.3 Solving for R 370
21.4.4 Finding π0 371
21.4.5 Performance Metrics 372
21.5 More Complex Chains 372
21.6 Readings and Further Remarks 376
21.7 Exercises 376
 contents xiii

22 Networks with Time-Sharing (PS) Servers (BCMP) 380

22.1 Review of Product-Form Networks 380
22.2 BCMP Result 380
22.2.1 Networks with FCFS Servers 381
22.2.2 Networks with PS Servers 382
22.3 M/M/1/PS 384
22.4 M/Cox/1/PS 385
22.5 Tandem Network of M/G/1/PS Servers 391
22.6 Network of PS Servers with Probabilistic Routing 393
22.7 Readings 394
22.8 Exercises 394
23 The M/G/1 Queue and the Inspection Paradox 395
23.1 The Inspection Paradox 395
23.2 The M/G/1 Queue and Its Analysis 396
23.3 Renewal-Reward Theory 399
23.4 Applying Renewal-Reward to Get Expected Excess 400
23.5 Back to the Inspection Paradox 402
23.6 Back to the M/G/1 Queue 403
23.7 Exercises 405
24 Task Assignment Policies for Server Farms 408
24.1 Task Assignment for FCFS Server Farms 410
24.2 Task Assignment for PS Server Farms 419
24.3 Optimal Server Farm Design 424
24.4 Readings and Further Follow-Up 428
24.5 Exercises 430
25 Transform Analysis 433
25.1 Definitions of Transforms and Some Examples 433
25.2 Getting Moments from Transforms: Peeling the Onion 436
25.3 Linearity of Transforms 439
25.4 Conditioning 441
25.5 Distribution of Response Time in an M/M/1 443
25.6 Combining Laplace and z-Transforms 444
25.7 More Results on Transforms 445
25.8 Readings 446
25.9 Exercises 446
26 M/G/1 Transform Analysis 450
26.1 The z-Transform of the Number in System 450
26.2 The Laplace Transform of Time in System 454
26.3 Readings 456
26.4 Exercises 456
27 Power Optimization Application 457
27.1 The Power Optimization Problem 457
27.2 Busy Period Analysis of M/G/1 459
27.3 M/G/1 with Setup Cost 462
xiv contents

27.4 Comparing ON/IDLE versus ON/OFF 465

27.5 Readings 467
27.6 Exercises 467

VII Smart Scheduling in the M/G/1

28 Performance Metrics 473
28.1 Traditional Metrics 473
28.2 Commonly Used Metrics for Single Queues 474
28.3 Today’s Trendy Metrics 474
28.4 Starvation/Fairness Metrics 475
28.5 Deriving Performance Metrics 476
28.6 Readings 477
29 Scheduling: Non-Preemptive, Non-Size-Based Policies 478
29.1 FCFS, LCFS, and RANDOM 478
29.2 Readings 481
29.3 Exercises 481
30 Scheduling: Preemptive, Non-Size-Based Policies 482
30.1 Processor-Sharing (PS) 482
30.1.1 Motivation behind PS 482
30.1.2 Ages of Jobs in the M/G/1/PS System 483
30.1.3 Response Time as a Function of Job Size 484
30.1.4 Intuition for PS Results 487
30.1.5 Implications of PS Results for Understanding FCFS 487
30.2 Preemptive-LCFS 488
30.3 FB Scheduling 490
30.4 Readings 495
30.5 Exercises 496
31 Scheduling: Non-Preemptive, Size-Based Policies 499
31.1 Priority Queueing 499
31.2 Non-Preemptive Priority 501
31.3 Shortest-Job-First (SJF) 504
31.4 The Problem with Non-Preemptive Policies 506
31.5 Exercises 507
32 Scheduling: Preemptive, Size-Based Policies 508
32.1 Motivation 508
32.2 Preemptive Priority Queueing 508
32.3 Preemptive-Shortest-Job-First (PSJF) 512
32.4 Transform Analysis of PSJF 514
32.5 Exercises 516
33 Scheduling: SRPT and Fairness 518
33.1 Shortest-Remaining-Processing-Time (SRPT) 518
33.2 Precise Derivation of SRPT Waiting Time∗ 521
 contents xv

33.3 Comparisons with Other Policies 523

33.3.1 Comparison with PSJF 523
33.3.2 SRPT versus FB 523
33.3.3 Comparison of All Scheduling Policies 524
33.4 Fairness of SRPT 525
33.5 Readings 529

Bibliography 531
Index 541
 Preface

The ad hoc World of Computer System Design

The design of computer systems is often viewed very much as an art rather than a
science. Decisions about which scheduling policy to use, how many servers to run,
what speed to operate each server at, and the like are often based on intuitions rather
than mathematically derived formulas. Specific policies built into kernels are often
riddled with secret “voodoo constants,”1 which have no explanation but seem to “work
well” under some benchmarked workloads. Computer systems students are often told
to first build the system and then make changes to the policies to improve system
performance, rather than first creating a formal model and design of the system on
paper to ensure the system meets performance goals.
Even when trying to evaluate the performance of an existing computer system, students
are encouraged to simulate the system and spend many days running their simulation
under different workloads waiting to see what happens. Given that the search space of
possible workloads and input parameters is often huge, vast numbers of simulations
are needed to properly cover the space. Despite this fact, mathematical models of the
system are rarely created, and we rarely characterize workloads stochastically. There is
no formal analysis of the parameter space under which the computer system is likely to
perform well versus that under which it is likely to perform poorly. It is no wonder that
computer systems students are left feeling that the whole process of system evaluation
and design is very ad hoc. As an example, consider the trial-and-error approach to
updating resource scheduling in the many versions of the Linux kernel.

Analytical Modeling for Computer Systems

But it does not have to be this way! These same systems designers could mathematically
model the system, stochastically characterize the workloads and performance goals,
and then analytically derive the performance of the system as a function of workload
and input parameters. The fields of analytical modeling and stochastic processes have
existed for close to a century, and they can be used to save systems designers huge
numbers of hours in trial and error while improving performance. Analytical modeling
can also be used in conjunction with simulation to help guide the simulation, reducing
the number of cases that need to be explored.

1 The term “voodoo constants” was coined by Prof. John Ousterhout during his lectures at the University of
California, Berkeley.

xvii
xviii preface

Unfortunately, of the hundreds of books written on stochastic processes, almost none

deal with computer systems. The examples in those books and the material covered are
oriented toward operations research areas such as manufacturing systems, or human
operators answering calls in a call center, or some assembly-line system with different
priority jobs.
In many ways the analysis used in designing manufacturing systems is not all that
different from computer systems. There are many parallels between a human operator
and a computer server: There are faster human operators and slower ones (just as
computer servers); the human servers sometimes get sick (just as computer servers
sometimes break down); when not needed, human operators can be sent home to save
money (just as computer servers can be turned off to save power); there is a startup
overhead to bringing back a human operator (just as there is a warmup cost to turning
on a computer server); and the list goes on.
However, there are also many differences between manufacturing systems and com-
puter systems. To start, computer systems workloads have been shown to have ex-
tremely high variability in job sizes (service requirements), with squared coefficients
of variation upward of 100. This is very different from the low-variability service times
characteristic of job sizes in manufacturing workloads. This difference in variability
can result in performance differences of orders of magnitude. Second, computer work-
loads are typically preemptible, and time-sharing (Processor-Sharing) of the CPU is
extremely common. By contrast, most manufacturing workloads are non-preemptive
(first-come-first-serve service order is the most common). Thus most books on stochas-
tic processes and queueing omit chapters on Processor-Sharing or more advanced pre-
emptive policies like Shortest-Remaining-Processing-Time, which are very much at
the heart of computer systems. Processor-Sharing is particularly relevant when analyz-
ing server farms, which, in the case of computer systems, are typically composed of
Processor-Sharing servers, not First-Come-First-Served ones. It is also relevant in any
computing application involving bandwidth being shared between users, which typi-
cally happens in a processor-sharing style, not first-come-first-serve order. Performance
metrics may also be different for computer systems as compared with manufacturing
systems (e.g., power usage, an important metric for computer systems, is not mentioned
in stochastic processes books). Closed-loop architectures, in which new jobs are not
created until existing jobs complete, and where the performance goal is to maximize
throughput, are very common in computer systems, but are often left out of queueing
books. Finally, the particular types of interactions that occur in disks, networking pro-
tocols, databases, memory controllers, and other computer systems are very different
from what has been analyzed in traditional queueing books.

The Goal of This Book

Many times I have walked into a fellow computer scientist’s office and was pleased to
find a queueing book on his shelf. Unfortunately, when questioned, my colleague was
quick to answer that he never uses the book because “The world doesn’t look like an
M/M/1 queue, and I can’t understand anything past that chapter.” The problem is that
Other documents randomly have
different content
Ronceval, thou noble valley!
Whensoe’er I hear thy name,
That blue flower so long departed
O’er my bosom sheds its fragrance!

Then the glitt’ring dream-world rises

Which for thousand years had faded,
And the mighty spirit-eyes
Gaze upon me, till I’m awe-struck!

Rattling sounds awake. There struggle

Saracen and Frankish knight;
As though bleeding and despairing
Ring Orlando’s bugle-notes

In the vale of Ronceval,

Hard beside Orlando’s gap—
Christen’d thus, because the hero,
Seeking how to force a passage,

With his trusty sword Duranda

Struck with such death-dealing fury
On the wall of rock, that plainly
To this day are seen its traces—

There within a gloomy hollow,

Close surrounded by a thicket
Of wild fir-trees, safely hidden,
Lies the cave of Atta Troll.

In the bosom of his fam’ly

Rests he after all the hardships
Of his flight and the distresses
Of his public show and travels.

Sweet the meeting! all his young ones

Found he in that happy cavern
Where with Mumma he begot them —
Where with Mumma he begot them,
Four his sons, and daughters two.

Well-lick’d maidens were the latter,

Fair their hair, like parsons’ daughters
Brown the youths, the youngest only
With the single ear is black.

Now this youngest was the darling

Of his mother, who when playing
Happen’d once to bite his ear off,
And for very love she ate it.

He’s a very genial stripling,

At gymnastics very clever,
And he turns a somersault
Like the posture-master Massmann.

Sprig of autochthonic humour,

He his mother-tongue loves only,
And has never learnt the jargon
Of the Grecian and the Roman.

Fresh and free and good and merry,

Soap he holds in detestation,
(Luxury of modern washing,)
Like the posture-master Massmann.

But our young friend is most genial

Where upon the tree he clambers,
Which along the steepest rock-side
From the deep abyss upriseth,

And extendeth to the summit,

When the family at night-time
Gather all around their father,
Toying in the evening coolness.

Then the old one loves to tell them

Wh h i h ld h i ’d
What he in the world has witness’d;
How he many men and cities
Had beheld, and greatly suffer’d,

Like Laertes’ noble offspring,

But in one thing still unlike him,—
Namely, that his wife went with him,
His dear black Penelope.

Atta Troll then also tells them

Of the wondrous approbation
That he, by his skill in dancing,
Had acquired in ev’ry quarter.

He assured them young and old

Had exultingly admired him,
When he danced upon the market
To the sweet notes of the bagpipe.

In particular the ladies,

Those dear connoisseurs of all things,
Had with vehemence applauded,
And had ogled him with favour.

O the vanity of Artists!

Our old dancing bear with simpers
Calls to mind the time when late he
To the public show’d his talent.

Overcome by self laudation,

He would fain by act exhibit
That he’s no mere boaster only,
But a really first-rate dancer.

From the ground then sudden springs he,

On his hinder paws upstanding,
And, as formerly, he dances
The gavotte, his favourite dance.
Mute, with muzzles gaping open,
The young bears look on with wonder,
While their father in the moonlight
Capers here and there thus strangely.

CAPUT V.
In the cavern, by his young ones,
Sick at heart, upon his back lies
Atta Troll, while thoughtful sucks he
At his paws, and sucks, and growls:

“Mumma, Mumma, swarthy jewel,

“Whom I out of life’s wide ocean
“Once did fish, in life’s wide ocean
“Once again I now have lost thee!

“Shall I ne’er again behold thee,

“Or beyond the grave p’rhaps only,
“Where, set free from earthly trammels,
“Thy dear soul is glorified?

“Would that I, alas! could once more

“Lick thy well-belovèd muzzle,
“My dear Mumma, which so sweetly
“Stroked me over, as with honey!

“Would that I again could snuffle

“That sweet smell, thy own peculiar,
“O my dear and swarthy Mumma,
“Charming as the scent of roses!

“But, alas! my Mumma’s pining

“In the fetters of those rascals,
“Who, the name of men adopting,
“Deem themselves creation’s masters.

“Death and hell! These men unworthy

“Aristocracy’s arch-emblems,
“Look down on the an’mal kingdom
“Proudly and disdainfully.

“Take away our wives and children,

“Fetter us, ill-treat us, even
“Kill us for the sake of selling
 Kill us, for the sake of selling
“Our poor hide and our poor carcass!

“And they think themselves permitted

“Wicked deeds like this to practise
“ ‘Gainst us bears especially,
“And the rights of man they call it!

“Rights of man indeed! Fine rights these.

“Tell me who bestow’d them on you?
“Nature certainly ne’er did so,
“For she’s not unnatural!

“Rights of man indeed! Who gave you

“This great privilege, I wonder?
“Reason certainly ne’er did so,
“For she’s not unreasonable!

“Men, pray are ye any better

“Than we others, just for eating
“All your dinners boil’d or roasted?
“In a raw state we eat ours,

“Yet is the result the same

“To us both.—No, food can never
“Make one noble; he is noble
“Who both nobly feels and acteth.

“Men, pray are ye any better

“Just because the arts and science
“With success ye follow? We now
“Never give ourselves the trouble.

“Are there not such things as learnèd

“Dogs, and horses too, who reckon
“Just like councillors of Commerce?
“Do not hares the drum play finely?

“Are not many beavers adepts

“I h fh d i ?
“In the art of hydrostatics?
“Were not clysters first invented
“By the cleverness of storks?

“Write not asses criticisms?

“Are not apes all good comedians?
“Is there any greater mimic
“Than Batavia, long tail’d monkey?

“Are not nightingales good singers?

“And is Freiligrath no poet,
“Who can sing of lions better
“Than his countryman the Camel?

“I myself the art of dancing

“Have advanced as much as Raumer
“That of writing. Writes he better
“Than I dance,—yes, I the bear?

“Men, why are ye any better

“Than we others? Upright hold ye,
“It is true, your heads, but in them
“Low-born thoughts are ever creeping.

“Men, pray are ye any better

“Than are we, because your skin is
“Smooth and glist’ning? This advantage
“Ye but share with every serpent.

“Human race, two leggèd serpents!

“Well I see the reason why ye
“Breeches wear; with foreign wool ye
“Hide your serpent-nakedness!

“Children, guard yourselves against these

“Hairless and misshapen creatures!
“My dear daughters, never marry
“Any monster that wears breeches!”
More than this I’ll not report now,
How the bear in his wild mania
For equality, kept reasoning
All about the human race.

For, to say the truth, I also

Am a man, and never will I
Tell again such foolish libels,
Which are, after all, offensive.

Yes, I am a man, and better

Than the other sucking creatures,
And the interests of the race
Ne’er will I renounce promoting.

In the fight with other creatures

Faithfully I’ll ever struggle
For humanity,—the holy
Rights of man that he is born to.

CAPUT VI.
Yet perchance ’tis beneficial
For us men, who form the higher
Kind of livestock, to discover
How they reason down below us.

Yes, below us, in the gloomy

Mournful spheres of fellowship,
In the beasts’ inferior strata,
Brood resentment, misery, pride.

That which natural hist’ry ever,

Equally with common custom,
Has for centuries admitted
Is denied with impious muzzle.

That false doctrine by the aged

In the young ones’ ears is grumbled
Which assails both cultivation
And humanity on earth.

“Children!” Atta Troll thus growl’d,

As he hither roll’d and thither
On his carpet-wanting couch:
“Unto us belongs the Future!

“If each bear but thought as I do,

“If all beasts but thought so too,
“With united forces would we
“Take up arms against the tyrants.

“Then the bear would form alliance

“With the horse, the elephant
“Twine his trunk in loving fashion
“Round the valiant ox’s horn.

“Bear and wolf of every colour,

“Goat and monkey, e’en the hare
“For a time would work in common
 For a time would work in common,
“And our triumph would be certain.

“Union, union is the’ essential

“Requisite; alone, we’re conquer’d
“Easily, but join’d together
“We would overreach the tyrants.

“Union! union! and we’ll triumph,

“And Monopoly’s vile sway
“Be o’erthrown, and we’ll establish
“A just kingdom for us beasts,

“Full equality for all, then,

“Of God’s creatures, irrespective
“Of their faith, or skin, or odour,
“Be its fundamental maxim!

“Strict equality! Each donkey

“Be entitled to high office;
“On the other hand, the lion
“Carry to the mill the sack.

“As respects the dog, indeed he

“Is a very servile rascal,
“Since for centuries has man
“Like a dog ne’er ceased to treat him.

“Yet in our free state we’ll give him

“Once again his olden rights,
“His prescriptive birthright, and he
“Soon again will be ennobled.

“Yes, the Jews shall then enjoy too

“All the rights of citizens,
“And by law be made the equals
“Of all other sucking creatures.

“Only dancing in the market

“F h J h ll b l f l
“For the Jew shall not be lawful;
“This amendment I insist on
“In the interest of my art.

“For a sense of style, of rigid

“Plastic art in motion’s wanting
“To that race, who really ruin
“What there is of public taste.”

CAPUT VII.
Gloomy, in his gloomy cavern,
Squats, in his belov’d home-circle,
Atta Troll, the misanthrope,
And he shows his teeth, and growls thus:

“Men, the pert and vulgar fellows!

“Smile away! From all your smiling
“And from your offensive yoke too
“Shall the coming day release us!

“I am always most offended

“By that sour-sweet kind of quiv’ring
“Round the mouth,—these smiles of man
“Find I really past all bearing!

“When I in his pallid visage

“See display’d that fatal quiv’ring,
“All my entrails in my body
“Turn right round with indignation.

“More impertinently even

“Than by words, a man lays open
“By his smile the deepest hidden
“Insolence of his vile spirit.

“They are always smiling! Even

“When by decency is needed
“Real solemnity of feature,—
“E’en in love’s most solemn moment!

“They are always smiling! Even

“When they’re dancing. In this manner
“They degrade this noble science,
“Which should be a kind of worship.

“Yes, the dance throughout all ages

“Was a pious act of faith;
“Solemnly around the altar
 Solemnly around the altar
“Turn’d the priests in mystic circle.

“Thus in olden time King David

“Danced before the ark of cov’nant;
“Dancing was an act of worship,
“Was a prayer upon the legs!

“I have ever understood thus

“Dancing, when upon the market
“To the people I was dancing,
“Who with their applause repaid me.

“This applause, I must confess it,

“Often made me feel quite happy;
“For extorting admiration
“From one’s foes is very sweet!

“But in their enthusiasm

“Still they smile. The art of dancing
“Powerless is to make them better,
“And they frivolous remain.”

CAPUT VIII.
Many a very virtuous burgher
Smells but badly, whilst the servants
Of a king with ambergris
Or else lavender are scented.

Virgin spirits may be met with

Which of green soap bear the odour,
Whilst the criminal with rose-oil
May have wash’d himself demurely.

Do not therefore turn your nose up,

Gentle reader, if the cave of
Atta Troll may not remind you
Of Arabia’s sweetest spices.

Tarry in that reeking circle,

’Mid those miserable stenches,
Where to his young son the hero
As from out a cloud thus speaks:

“Child, my child, thou youngest offspring

“Of my loins, now place thy one ear
“Close beside thy father’s muzzle,
“And suck in my solemn words!

“Guard against man’s ways of thinking,

“They destroy both soul and body;
“ ‘Mongst all men there’s no such thing as
“Any ordinary man.

“E’en the Germans, once so noble,

“E’en the very sons of Tuisco,
“Our own primitive relations,
“They too have degenerated.

“They’ve become now faithless, godless,

“Even preaching atheism—
“Child my child be on thy guard
 Child, my child, be on thy guard,
“ ‘Gainst both Feuerbach and Bauer![30]

“Never be an Atheist,
“Monster void of all respect for
“The Creator—a Creator
“ ’Twas who made this universe!

“High above us, sun and moon

“And the stars too (both the tail-less
“And all those with tails provided)
“Are reflections of His power.

“Down below us, land and sea

“Are the echo of His glory,
“And each living creature praises
“Evermore His excellencies.

“E’en the smallest silver-louse that

“In the aged pilgrim’s beard
“In life’s pilgrimage is sharer,
“Sings the great Eternal’s praises!

“In yon starry bright pavilion,

“On the golden seat of power,
“World-directing and majestic,
“Sits a mighty polar bear.

“Free from spot and snow-white glitt’ring

“Is his skin; his head is cover’d
“With a crown of diamonds,
“Which illumines all the heavens.

“In his face is harmony,

“And the silent deeds of thinking;
“If he signs but with his sceptre,
“All the spheres resound with singing.

“At his feet bear-saints are sitting

“Piously, who meekly suffer’d
“While on earth, and in their paws they
“Hold the palms of martyrdom.

“Ofttimes one amongst them rises,

“Then another,—by the Spirit
“Seeming mov’d, and straightway dance they
“Their most solemn sacred dance—

“Sacred dance, where mercy’s radiance

“Renders talent quite superfluous,
“And the soul for very rapture
“From the skin attempts to leap!

“O shall I, unworthy Troll,

“E’er partake this great salvation?
“And from earth’s debasing sorrows
“To the realms of bliss soar upwards?

“O shall I, all-drunk with heaven,

“In the stars’ pavilion yonder,
“With the palm and with the glory,
“Dance before the Master’s throne?”

CAPUT IX.
Like the tongue as red as scarlet,
Which a swarthy Freiligrathian
Moorish prince with scornful fury
From his sullen mouth protruded,

So the moon from out the gloomy

Clouds of heaven advanced. Afar off
Cataracts are roaring, sleepless
And morosely through the night.

Atta Troll upon the summit

Of his fav’rite rock stands lonely,
Lonely, and to the abyss
Downward howls he in the nightwind:

“Yes, I am a bear, I am so,—

“Him ye christen shaggy bear,
“Growler, Isegrim, and Bruin,
“And heav’n knows how many others.

“Yes, I am a bear, I am so,

“The uncouth and boorish creature,
“I’m the awkward dromedary
“Of your scorn and cruel laughter.

“I’m the butt of all your wit,

“I’m the bugbear, with whose terrors
“Ye at night your children frighten,
“Human children, when they’re naughty.

“I’m the joke of all your idle

“Nurs’ry stories, well I know it,
“And I now proclaim it loudly
“To man’s paltry world below.

“Hear it, hear; a bear am I,

“My descent I’m not ashamed of,
“But am proud of it as though I
 But am proud of it, as though I
“Sprang from Moses Mendelssohn!”

CAPUT X.
Two dark figures, wild and surly,
And upon their all-fours gliding,
Force their way across the gloomy
Grove of firs at midnight’s hour.

This is Atta Troll, the father,

And his son, young master one-ear.
Where the wood grows somewhat lighter
By the stone of blood they halted.

“This old stone”—growl’d Atta Troll,—

“Is the altar where the Druids
“In the days of superstition
“Human sacrifices offer’d.

“O their cruelty accursèd!

“All the hair upon my back
“Bristles when I think upon it;
“Blood was pour’d out to God’s honour!

“Now these men are more enlighten’d,

“And no longer kill each other
“Merely in excessive zeal
“For the interests of heaven.

“ ’Tis no longer pious fancies,

“Madness, nor enthusiasm,
“But mere vanity and self-love
“Makes them now commit their murders.

“On the good things of the earth

“Eagerly they’re ever seizing;
“ ’Tis an endless round of fighting,
“For himself each person stealeth!

“Yes! the heritage of all

“Is the individual’s booty;
“Of the rights then of possession
 Of the rights, then, of possession
“Speaks he, thinking of his own!

“Of his own! Possession’s rights too!

“O, the cruel theft, the lying!
“None but man could have invented
“Such commingled fraud and madness.

“Private property was never

“Made by Nature; pocketless,
“With no pockets in our skins, we
“Ev’ry one the world first entered.

“Not a single one amongst us

“At his birth had such a pocket
“In his body’s outer skin,
“Where he might conceal his robb’ries.

“Man alone, that smooth-skinn’d being,

“Who with foreign wool so nicely
“Clothes himself, had e’er the sharpness
“To provide himself with pockets.

“Pockets! They’re as much ’gainst nature

“As is private property,
“As possession’s rights themselves are—
“Men in fact are but pickpockets!

“Fiercely hate I them! My hatred

“Unto thee, my son, bequeath I;
“Here upon this altar shalt thou
“Swear to man undying hatred!

“Be implacably the death-foe

“Of those wicked vile oppressors
“To the very end of life,—
“Swear it, swear it here, my son!”

And the youngster swore, as once did

H ib l Th ll ll
Hannibal. The moon, all yellow,
On the stone of blood look’d wildly,
And the pair of misanthropes.

By-and-by we’ll tell the story

How the young bear ever faithful
To his oath remain’d. Our lyre shall
In another Epic praise him.

As respects friend Atta Troll,

We will leave him for the present,
Presently to come across him,
All the surer, with a bullet.

All thy stealthy machinations,

Traitor ’gainst man’s majesty,
Now at length are terminated,
And thy hour will sound to-morrow!

CAPUT XI.
Like some drowsy bayaderes
Look the mountains, standing shiv’ring
In their snowy shirts of clouds,
Flutt’ring in the breeze of morning.

Yet they soon become enliven’d

By the sun-god stripping from them
All the veil that’s hanging o’er them
Lighting up their naked beauty!

Early in the morn I started

With Lascaro on our journey
Bound to hunt the bear. At noonday
We arrived at Pont d’Espagne.

So they call the bridge which leadeth

Out of France and into Spain,
To the land of west barbarians,
Who’re a thousand years behind us,—

Yes, a thousand years behind us

In all modern civ’lisation;
My barbarians to the eastward
But a hundred years behind are.

Slowly, almost trembling, left I

France’s sacred territory,
Blessèd fatherland of freedom
And the women that I love!

On the middle of the bridge

A poor Spaniard sat. Deep mis’ry
Lurk’d behind his tatter’d mantle,
Misery in his eyes was lurking.

An old crazy mandoline

With his wither’d fingers pinch’d he;
Shrill the discord which re-echoed
Shrill the discord which re-echoed
From the rocks, as in derision.

Oftentimes his figure bent he

Downward tow’rd the’ abyss with laughter,
Tinkling harder then than ever,
While the following words he sang:

“In the middle of my bosom

“Stands a little golden table;
“Round the little golden table
“Stand four little golden chairs.

“On the golden chairs are sitting

“Little ladies, golden arrows
“In their hair,—at cards they’re playing,
“But ’tis only Clara wins.

“As she wins, she laughs with slyness;

“Ah! within my bosom, Clara,
“Thou’lt be ev’ry time a winner,
“For thou holdest nought but trumps.”

Wand’ring onward, to myself I

Spoke: “ ’Tis singular that madness
Sits and sings upon yon bridge,
That from France to Spain leads over.

“Is this madman but the emblem

“Of the interchange ’mongst nations
“Of their thoughts? or his own country’s
“Wild and crazy title-page?”

We arrived not until evening

At the wretched small posada,
Where an olla-podrida
In a dirty dish was smoking.

There I swallow’d some garbanzos,

H l k b ll
Heavy, large as musket-bullets,
Indigestible to Germans,
Though to dumplings they’re accustom’d.

Fit companion to the cooking

Was the bed. With insects pepper’d
It appear’d. The bugs, alas! are
Far the greatest foes of man.

Fiercer than the wrath of thousand

Elephants, I find the hatred
Of one tiny little bug,
When across my bed it crawleth.

One must let them bite in quiet,—

This is bad enough,—still more ’tis
If one crushes them. The stink then
Keeps one all night long in torment.

Yes, the fiercest earthly trouble

Is the fight with noxious vermin,
Who a stench employ as weapons,—
Is a duel with a bug!

CAPUT XII.
How they rave, the race of poets,
E’en the tame ones, singing ever
And exclaiming: “Nature’s surely
“The Creator’s mighty temple—

“Is a temple all whose glories

“To our Maker’s fame bear witness,
“Sun and moon and stars all hanging
“In its cupola as lamps.”

Well and good, my worthy people!

Yet confess that in this temple
Are the stairs uncomfortable,
Bad and inconvenient stairs!

All this up-and-down-stairs going,

Mountain-climbing and this jumping
Over rocks is very tiring
To the legs as well as spirit.

Close beside me walk’d Lascaro,

Pale and lanky, like a taper;
Never spoke he, never laugh’d he,
He, the dead son of the sorc’ress.

Yes, ’tis said that he’s a dead man,

Dead long since, but yet his mother
Old Uraca’s magic science
Kept him living in appearance.—

That accursèd temple-staircase!

It exceeds my comprehension
How my neck escaped from breaking,
Stumbling o’er a precipice.

How the cataracts were shrieking!

How the tempest flogg’d the fir-trees
Till they howl’d! The clouds began too
Till they howl d! The clouds began too
Crashing suddenly—bad weather!

In a little fishing cottage

By the Lac-de-Gobe soon found we
Shelter and some trout for luncheon;
Most delicious were the latter.

In an arm-chair was reclining,

Ill and grey, the ferryman;
On him his two pretty nieces,
Like a pair of angels, waited.

Stoutish angels, rather Flemish,

Seeming from a frame descended
Of a Rubens; gold their tresses,
Full of health their eyes, and liquid.

Their vermilion cheeks were dimpled,

With a secret slyness in them;
Strong their limbs were, and voluptuous,
Giving pleasure to the fancy.

Dear, affectionate young creatures,

Keeping up a sweet discussion,
As to which drink would be relish’d
Most of all by their sick uncle.

If the one the cup should bring him

Full of well-boil’d linden blossoms,
Then the other hastes to feed him
With an elder-flow’r decoction.

“I’ll not drink of either of them,”

“Cried impatiently the old man;
“Fetch some wine, that I may offer
“To my guests some better drink!”

Whether it was wine they gave me

A h L d G b I ll
At the Lac-de-Gobe, I really
Cannot say. Methinks in Brunswick
By the name of Mum they’d call it.

Of the very best black goat-skin

Was the wine-skin, stinking foully;
Yet the old man drank with pleasure,
And he seem’d quite well and joyous.

He recounted the achievements

Of the smugglers and banditti
Merrily and freely living
In the Pyrenean forests.

Many old traditions also

Well he knew: amongst the others
Were the battles of the giants
With the bears in times primeval.

Yes, the bears then and the giants

Struggled fiercely for the mast’ry
Of these mountains and these valleys,
Ere by man they were discover’d.

But when man arrived, the giants

Fled away from out the country
Stupified, for little brains
Are contain’d in heads gigantic.

And ’tis said the silly fellows,

On arriving at the ocean,
And observing how the heavens
In its azure depths were mirror’d,

Cleverly supposed the ocean

To be heaven, and plunged down in it,
Full of godlike confidence,
And were drown’d, the whole together
As respects the bears, however,
They are gradually being
Kill’d by man, their numbers yearly
In the mountain still decreasing.

“Thus on earth” exclaim’d the old man,

“One gives place unto another,
“And when men are put an end to,
“Then the dwarfs will be the masters.

“Yes, the clever little people,

“Who the mountain’s womb inhabit,
“ ‘Mongst the golden mines of riches
“Digging and collecting nimbly.

“How they from their hiding-places

“With their small sly heads keep peeping!
“Oft I’ve seen them in the moonlight,
“And then trembled at the future;

“At the power their gold will give them;

“Ah, I fear lest our descendants
“Fly for refuge, like the stupid
“Giants, to the watery heaven!”

CAPUT XIII.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookultra.com