Professional Documents
Culture Documents
Machine Learning For Cloud Management 1St Edition Kumar Online Ebook Texxtbook Full Chapter PDF
Machine Learning For Cloud Management 1St Edition Kumar Online Ebook Texxtbook Full Chapter PDF
https://ebookmeta.com/product/machine-learning-techniques-and-
analytics-for-cloud-security-advances-in-learning-analytics-for-
intelligent-cloud-iot-systems-1st-edition-chakraborty/
https://ebookmeta.com/product/machine-learning-for-intelligent-
multimedia-analytics-techniques-and-applications-pardeep-kumar-
amit-kumar-singh-eds/
https://ebookmeta.com/product/machine-learning-big-data-and-iot-
for-medical-informatics-1st-edition-pardeep-kumar/
https://ebookmeta.com/product/prediction-and-analysis-for-
knowledge-representation-and-machine-learning-1st-edition-
avadhesh-kumar-editor/
Machine Learning Approach for Cloud Data Analytics in
IoT 1st Edition Sachi Nandan Mohanty (Editor)
https://ebookmeta.com/product/machine-learning-approach-for-
cloud-data-analytics-in-iot-1st-edition-sachi-nandan-mohanty-
editor/
https://ebookmeta.com/product/beginning-machine-learning-in-the-
browser-nagender-kumar-suryadevara/
https://ebookmeta.com/product/machine-learning-and-optimization-
techniques-for-automotive-cyber-physical-systems-1st-edition-
vipin-kumar-kukkala/
https://ebookmeta.com/product/artificial-intelligence-and-
machine-learning-in-business-management-concepts-challenges-and-
case-studies-1st-edition-sandeep-kumar-panda-editor/
https://ebookmeta.com/product/machine-learning-for-financial-
risk-management-with-python-algorithms-for-modeling-risk-1st-
edition-karasan/
Machine Learning for
Cloud Management
Machine Learning for
Cloud Management
Jitendra Kumar
Ashutosh Kumar Singh
Anand Mohan
Rajkumar Buyya
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742
© 2022 Jitendra Kumar, Ashutosh Kumar Singh, Anand Mohan, Rajkumar Buyya
Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot as-
sume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have
attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders
if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please
write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or
utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including pho-
tocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission
from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.com or contact the
Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that are
not available on CCC please contact [email protected]
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are used only for iden-
tification and explanation without intent to infringe.
DOI: 10.1201/9781003110101
Publisher’s note: This book has been prepared from camera-ready copy provided by the authors.
Dedicated to,
My wife: Gita, daughter: Aru, and Parents
∼Jitendra Kumar
∼Anand Mohan
∼Rajkumar Buyya
Contents
List of Figures xi
Preface xix
Author xxi
Abbreviations xxv
Chapter 1 Introduction 1
2.1 AUTOREGRESSION 14
2.2 MOVING AVERAGE 14
2.3 AUTOREGRESSIVE MOVING AVERAGE 15
vii
viii Contents
Bibliography 159
Index 171
List of Figures
xi
xii LIST OF FIGURES
3.1 Wilcoxon test statistics for error preventive and non-error preventive
time series forecasting model 57
3.2 Finner test post-hoc analysis of error preventive and non-error pre-
ventive time series forecasting models 57
xvii
xviii LIST OF TABLES
7.1 ARIMA analysis orders for cloud resource demand traces 124
7.2 Network configuration parameter choices for decomposition predictive
framework 124
7.3 List of experiments selected by D-Optimal Design 124
7.4 Mean squared error of short term forecast of ELM based predictive
framework 134
7.5 Mean squared error of long term forecast of ELM based predictive
framework 137
7.6 Forecast accuracy comparison of ELM based predictive models on
CPU trace with state-of-art models 138
7.7 Forecast accuracy comparison of ELM based predictive models on
Memory trace with state-of-art models 138
7.8 Forecast accuracy comparison of ELM based predictive models on
Google cluster trace and PlanetLab Trace with state-of-art models 139
Cloud computing has become one of the revolutionary technology in the history
of the computing world. It offers subscription-based on-demand services and has
emerged as the backbone of the computing industry. It has enabled us to share
resources among multiple users through virtualization by the means of creating a
virtual instance of a computer system running in an abstracted hardware layer. Unlike
early distributed computing models, it assures limitless computing resources through
its large-scale cloud data centers. It has gained wide popularity over the past few
years, with an ever-increasing infrastructure, number of users, and amount of hosted
data. The large and complex workloads hosted on these data centers introduce several
challenges: resource utilization, power consumption, scalability, operational cost, and
many others. Therefore, a practical resource management scheme is essential to bring
operational efficiency with improved elasticity. The elasticity of a system depends on
several factors such as the accuracy of anticipated workload information, performance
behavior of applications in different scenarios communicating the forecast results, use
of the anticipated information, and many others.
Effective resource management can be achieved through workload prediction,
resource scheduling, and provisioning, virtual machine placement, or a combination
of these approaches. The workload prediction has been widely explored and a number
of methods are presented. However, the existing methods suffer from various issues
including the incapability of capturing the non-linearity of workloads and iterative
training that consumes huge computing resources and time. This book discusses
the machine learning-based approaches to address the above-mentioned issues. The
highlights of the discussed models are continuous learning from error feedback, adaptive
nature, decomposition of workload traces, and ensemble learning. Detailed analysis of
predictive methods on different workload traces is also included and their performance
is compared with state-of-art models. Virtual machine placement is another aspect
that is explored to achieve efficient resource management. In general, virtual machine
placement is a multiobjective problem that involves multiple conflicting objectives to
be optimized simultaneously. The frameworks discussed in this book address the issues
of resource utilization, power consumption, and security while placing the workloads
on servers.
The remainder of the book is organized as follows: Chapter 1 briefs the basic cloud
computing concepts. The discussion on the workload prediction mechanisms begins
in chapter 2. First, the basic time series forecasting models are discussed with their
performance on different workload traces. Chapter 3 discusses the error preventive time
series forecasting models which significantly improve the performance over classical
time series models. Then, a discussion on various nature-inspired algorithms is included
xix
xx Preface
xxi
xxii Author
Indian Science Congress held in Chennai, INDIA, Best Paper Presenter of NSC’99
INDIA and Bintulu Development Authority Best Postgraduate Research Paper Award
for 2010, 2011, 2012.
He has served as an Editorial Board Member of International Journal of Networks
and Mobile Technologies, International Journal of Digital Content Technology and
its Applications. Also, he has shared his experience as a Guest Editor for Pertanika
Journal of Science and Technology, Chairman of CUTSE International Conference
2011, Conference Chair of series of International Conference on Smart Computing and
Communication (ICSCC), and as an editorial board member of UNITAR e-journal.
He is involved in reviewing processes in different journals and conferences of repute
including IEEE transaction of computer, IET, IEEE conference on ITC, ADCOM,
etc.
Prof. Anand Mohan has nearly 44 years of experience in teaching and research
and the administration and management of higher educational institutions. He is
currently an institute professor in the Department of Electronics Engineering, Indian
Institute of Technology (BHU), Varanasi, India. Besides his present academic assign-
ment, Prof. Mohan is a Member of the Executive Council of Banaras Hindu University
and Vice-Chairman of the Board of Governors of Indian Institute of Technology
(BHU), Varanasi, India. Prof. Mohan served as Director (June 2011-June 2016) of
the National Institute of Technology (NIT), Kurukshetra, Haryana, India, and was
also Mentor Director of the National Institute of Technology, Srinagar, Uttarakhand,
India. For his outstanding contributions in the field of Electronics Engineering, Prof.
Mohan was conferred the ’’Lifetime Achievement Award’’ (2016) by Kamla Nehru
Institute of Technology, Sultanpur, India.
Prof. Rajkumar Buyya is a Redmond Barry Distinguished Professor and Direc-
tor of the Cloud Computing and Distributed Systems (CLOUDS) Laboratory at the
University of Melbourne, Australia. He is also serving as the founding CEO of Manjra-
soft Pty Ltd., a spin-off company of the University, commercializing its innovations in
Cloud Computing. He served as a Future Fellow of the Australian Research Council
during 2012-2016. He serving/served as an Honorary/Visiting Professor for several
elite Universities including Imperial College London (UK), University of Birmingham
(UK), University of Hyderabad (India), and Tsinghua University (China). He received
B.E and M.E in Computer Science and Engineering from Mysore and Bangalore
Universities in 1992 and 1995 respectively; and a Doctor of Philosophy (Ph.D.) in
Computer Science and Software Engineering from Monash University, Melbourne,
Australia in 2002. He was awarded Dharma Ratnakara Memorial Trust Gold Medal
in 1992 for his academic excellence at the University of Mysore, India. He received
Richard Merwin Award from the IEEE Computer Society (USA) for excellence in
academic achievement and professional efforts in 1999. He received Leadership and
Service Excellence Awards from the IEEE/ACM International Conference on High-
Performance Computing in 2000 and 2003. He received the ‘‘Research Excellence
Awards’’ from the University of Melbourne for productive and quality research in
computer science and software engineering in 2005 and 2008. With over 112,400
citations, a g-index of 322, and an h-index of 145, he is the highest cited computer
scientist in Australia and one of the world’s Top 30 cited authors in computer science
Author xxiii
and software engineering. He received the Chris Wallace Award for Outstanding
Research Contribution 2008 from the Computing Research and Education Association
of Australasia, CORE, which is an association of university departments of computer
science in Australia and New Zealand. Dr. Buyya received the ‘‘2009 IEEE TCSC
Medal for Excellence in Scalable Computing’’ for pioneering the economic paradigm for
utility-oriented distributed computing platforms such as Grids and Clouds. He served
as the founding Editor-in-Chief (EiC) of IEEE Transactions on Cloud Computing
(TCC). Dr. Buyya is recognized as a ‘‘Web of Science Highly Cited Researcher’’ for
five consecutive years since 2016, a Fellow of IEEE and Scopus Researcher of the
Year 2017 with Excellence in Innovative Research Award by Elsevier, and ‘‘Lifetime
Achievement Award’’ from two Indian universities for his outstanding contributions
to Cloud computing and distributed systems. He has been recently recognized as the
‘‘Best of the Worl’’, in the Computing Systems field, by The Australian 2019 Research
Review.
Dr. Buyya has authored/co-authored over 850 publications. Since 2007, he received
twelve ‘‘Best Paper Awards’’ from international conferences/journals including a
‘‘2009 Outstanding Journal Paper Award’’ from the IEEE Communications Society,
USA. He has co-authored five text books: Microprocessor x86 Programming (BPB
Press, New Delhi, India, 1995), Mastering C++ (McGraw Hill Press, India, 1st
edition in 1997 and 2nd edition in 2013), Object Oriented Programming with Java:
Essentials and Applications (McGraw Hill, India, 2009), Mastering Cloud Computing
(Morgan Kaufmann, USA; McGraw Hill, India, 2013; China Machine Press, 2015),
and Cloud Data Centers and Cost Modeling (Morgan Kaufmann, USA, 2015). The
books on emerging topics that he edited include High Performance Cluster Computing
(Prentice Hall, USA, 1999), High Performance Mass Storage and Parallel I/O (IEEE
and Wiley Press, USA, 2001), Content Delivery Networks (Springer, Germany, 2008),
Market Oriented Grid and Utility Computing (Wiley Press, USA, 2009), and Cloud
Computing: Principles and Paradigms (Wiley, USA, 2011). He also edited proceedings
of over 25 international conferences published by prestigious organizations, namely
the IEEE Computer Society Press (USA) and Springer Verlag (Germany). He served
as Associate Editor of Elsevier’s Future Generation Computer Systems Journal (2004-
2009) and currently serving on editorial boards of many journals including Software:
Practice and Experience (Wiley Press). Dr. Buyya served as a speaker in the IEEE
Computer Society Chapter Tutorials Program (from 1999-2001), Founding Co-Chair
of the IEEE Task Force on Cluster Computing (TFCC) from 1999-2004, and member
of the Executive Committee of the IEEE Technical Committee on Parallel Processing
(TCPP) from 2003-2011. He served as the first elected Chair of the IEEE Technical
Committee on Scalable Computing (TCSC) during 2005-2007 and played a prominent
role in the creation and execution of several innovative community programs that
propelled TCSC into one of the most successful TCs within the IEEE Computer
Society. In recognition of these dedicated services to the computing community
over a decade, the President of the IEEE Computer Society presented Dr. Buyya a
Distinguished Service Award in 2008.
Dr. Buyya has contributed to the creation of high-performance computing and
communication system software for PARAM supercomputers developed by the Centre
xxiv Author
xxv
xxvi Abbreviations
Introduction
DOI: 10.1201/9781003110101-1 1
2 Machine Learning for Cloud Management
SaaS
User Ap-
plications
PaaS
Database Application
UI Services
Grid Grid
IaaS
disaster recovery, accessibility, measured services, and many others [25, 73, 75, 98].
However, various challenges and limitations are still open including inefficient resource
management, security and privacy, heterogeneity, elasticity, usability, response time,
and many more [18, 19, 21, 52--54, 90, 109, 124, 125].
Resource
Clients Information
Web
Portal
Resource Management
Resource Pool
Workload
Workload Resource
Analysis Provision
Cloud Scheduler
a modern cloud system tries to assign the workloads to one of the server machines
based on different criteria including resource utilization, system performance, user
priorities, operational cost, quality of service, etc. Typically, the complete process of
workload placement over a time to improve different variables of a system is referred
to as cloud resource management. As depicted in Fig. 1.2, the major tasks of a cloud
resource management application are workload analysis and forecasting, resource
provisioning, and scheduling the workloads on hardware. The workload analysis
module is responsible for analyzing the upcoming workload and for forecasting the
expected workload in the near future. This information is used by the resource
provisioning module to allocate the physical resources. The resource scheduler places
the workloads on the servers based on the input from the resource provisioning module
and current resource usage information. Typically, resource management is achieved
through prediction, scaling, provisioning, and load balancing, as shown in Fig. 1.3.
However, this book concentrates on workload forecasting using different approaches
of regression analysis and artificial neural networks, and load balancing.
Cloud Resource
Management
Forecaster
Cloud Datacenter
Se Se
rve rve
r r
Se Se
rve rv
r er
WWW Load
Balancer
Cloud Users
The effective load balancing is another approach that helps in achieving better
usage of resources and their management. The efficiency of load balancing approaches
has been an issue for cloud systems since its development [84, 129]. The efficiency in
load balancing can be achieved using different approaches such as optimal scheduling
and placement of workloads or virtual machines (VMs). The optimal mapping of
VMs is a complex and challenging task as it involves multiple objectives to optimize
at the same time and belongs to NP-Complete class of problems [16, 88]. Generally,
the existing VM placement algorithms consider the different dimensions of resource
utilization and power consumption in the data centers [4, 138]. We will focus on the
load balancing approaches, also dealing with the security while balancing the load on
cloud servers as it is one of the most important issues in the cloud architectures, and
various approaches have been discovered including [43, 85].
Output
Layer
Hidden Hidden
Layer-1 Layer-2
The key difference between a traditional computing approach and a neural network
is that the traditional approach follows a set of rules that must be known to the
computer in advance, while a neural network can learn from the data itself to draw
insightful inferences using some specific rules. Let κ1 = [x1 , x2 , . . . , xt ] be an input
vector, and the network, as shown in Fig. 1.6, is applied to estimate the value of xt+1 .
Assuming that ωi,j k
represents the weight of a synaptic connection between the ith
node of the k th layer and the j th node of the next layer, and ζk denotes the activation
function applied on k th layer nodes. The output of the j th node of layer k + 1 can be
computed as zj = ti=1 ζk+1 (xi · ωi,jk
) that acts as the input to next layer nodes.
P
farb
These algorithms can be classified into two major categories i.e. trajectory-based
and population-based approaches. A trajectory-based algorithm such as Simulated
Annealing works around a single solution to find an optimal solution for the problem
under consideration. On the other hand, a population-based algorithm uses a set
of solutions to search for an optimal solution. A detailed study on metaheuristic
optimization can be seen in [14].
HTTP-Web Server Logs: The HTTP traces of web servers of NASA, Calgary, and
Saskatchewan servers are used [1]. In this book, these data traces are referred to as
NASA Trace (D1 ), Calgary Trace (D2 ), Saskatchewan Trace (D3 ), respectively. The
D1 is composed of two months of HTTP web requests obtained from the WWW server
of NASA Kennedy Space Center in Florida. Similarly, the D2 data-trace contains the
HTTP request of one-year duration obtained from the WWW server located at the
University of Calgary, Alberta, Canada. On the other hand, the D3 is a data-trace
that contains the HTTP server requests of seven months obtained from a WWW
8 Machine Learning for Cloud Management
server of a university at Saskatchewan. Every data trace stores the records in ASCII
files, and every line stores one record. Every record is composed of five records i.e.
host, timestamp, request, HTTP reply code, and bytes in the reply.
Google Cluster Trace: It contains the data collected from the cluster cell of Google
for 29 days of duration. The workload trace was released in 2011, and it contains
the data from 10388 servers, 20 million tasks, and more than 0.67 million jobs [112].
A job is a set of one or more tasks, and tasks are further decomposed into one or
more processes. In this book, the CPU and Memory resource demands are used and
referred to as CPU Trace (D4 ) and Memory Trace (D5 ).
Mean Squared Error: The mean squared error (MSE) measures the forecast accuracy,
and it is one of the popular metrics used in the literature. This method heavily penalizes
the large error terms. Mathematically, it is denoted as given in eq. (1.2), where m
represents the size of data in a given trace. The term MSE and MPE (mean squared
prediction error) are interchangeably used in the book. Moreover, the square root of
MSE (RMSE) may also be used as an error metric.
m
1 X
M SE = (xt − x̂t )2 (1.2)
m t=1
Mean Absolute Error: A small number of very large magnitude errors may influence
the accuracy measured using mean squared error. Whereas mean absolute error equally
weights every error term, and it computes the mean of absolute differences between
predicted and actual workloads as given in eq. (1.3). The forecasts are close to the
actual workload values if the measured score is close to zero.
m
1 X
M AE = |xt − x̂t | (1.3)
m t=1
Introduction 9
Relative Mean Absolute Error: A scale-free error metric is required to compare the
forecast models on different data sets, and relative mean absolution error (RelMAE)
is one such metric. The score can be calculated using eq. (1.4), which represents the
mean absolute error of the algorithm (M AEA ) normalized by the mean absolute error
of a base or state of the art model (M AEBM )
M AEA
RelM AE = (1.4)
M AEBM
Mean Absolute Scaled Error: Rob J. Hyndman and Anne B. Koehler introduced a
new metric as a substitution of percentage error metrics [61]. The prediction errors
are scaled on the basis of the training mean absolute error of a naı̈ve forecast method.
It computes the measured score using eq. (1.5), where ms denotes the seasonal term.
This metric is a good choice of accuracy measurement when the prediction model is
compared across a number of different scales.
m
1 X
!
|xt − x̂t |
M ASE(x, x̂) = 1 Pm (1.5)
m t=1 m−ms t=ms +1 |xt − xt−1 |
Sum of Elasticity Index: Messias et al. proposed to use the sum of elasticity index
(SEI) as a measure of forecast accuracy [92]. This metric supports a forecast model
having the best performance most of the time. As opposed to MAE and RMSE, it is
very less sensitive to the outliers. The SEI score is computed as given in eq. (1.7) and
it always lies between zero and one, where zero and one define the worst and best
accuracy of the model.
m
min(xt , x̂t )
SEI = (1.7)
X
t=1
max(xt , x̂t )
i >0
2 i =0
1X
WC =
R− rank(i ) + rank(i ) (1.9)
X
i <0
2 i =0
significance level (ℵ) in a step-down manner [34]. Considering that the generated
p-values are sorted in an increasing fashion in such a way that pi ≤ pi+1 ; ∀i =
{1, 2, . . . , k − 2}. Let HiFN be the corresponding hypothesis for tests. The Finner test
rejects the hypothesis from H1FN to Hi−1 FN
provided i is the smallest integer number
k−1
that satisfy pi > 1 − (1 − ℵ) i property [34].
CHAPTER 2
ime series analytical models are being used in forecasting since a long ago in
T 1927 [128]. A time series-based model forecasts the trends after analyzing the
various characteristics of data indexed in time. Since their first usage, they have been
widely used in scientific research and industry-oriented applications. This chapter
concentrates on univariate time series-based workload forecasting. A univariate time
series can be defined as a collection of measurements of the same variable over time
(typically, at regular time intervals). The essential characteristic of any time series
data is that the order of observation matters and change in order may alter the
significance of the data.
The time series analysis is typically associated with the process of finding a model
to fit the time series data. The observed model can be used to extract the pattern,
forecast future events, and explain the effects of past events on the future. Some of
the essential characteristics of a time series are:
• The trend depicts the direction of the data i.e. to increase or decrease. The
direction is always need not be in the same direction for a long period of time.
According to the Organisation for Economic Co-operation and Development
(OECD), ‘‘The trend is the component of a time series that represents variations
of low frequency in a time series, the high and medium frequency fluctuations
having been filtered out.’’
• The seasonality also depicts similar characteristics as of trend. The difference
between the two terms is that the seasonality shows repetitive patterns.
• The noise referred to the component depicting neither trend nor seasonality in
the data.
• Outliers are the far-away data points from the data.
• The other common characteristics in a time series data are long-run cycle,
constant variance over time, and spikes.
This chapter discusses the five basic time series analysis models and uses them
to forecast the different types of workloads on cloud servers. A detailed analysis is
conducted to validate their performance on real-world data traces.
DOI: 10.1201/9781003110101-2 13
14 Machine Learning for Cloud Management
2.1 AUTOREGRESSION
An autoregressive model is used to present a phenomenon where the future values of
any variable are the function of its historical values. It is used to depict a random
process in real-world applications of statistics, signal processing, data analysis, etc.
Formally, an autoregressive model can be defined as a process that considers the
historical data with a white noise term to generate the future outcome of a variable.
This model gets the name from its functioning as it regresses the same variable.
Considering that the cloud workload x is indexed in equally spaced time interval
1, 2, . . . , t as x1 , x2 , . . . , xt then the autoregressive model of order p can be defined as
eq. (2.1) [91].
φ2 φ1
φp
Prediction
fAR
ℵt
θ2 θ1
θq
Prediction
fMA
ℵt
x̂t = φ1 ×xt−1 +φ2 ×xt−2 +. . .+φp ×xt−p +ℵt +θ1 ×ξt−1 +θ2 ×ξt−2 +. . .+θq ×ξt−q (2.3)
φp φ2
φ1
Prediction
ℵt fARMA x̂t
φp φ1
φ2
process is a good choice to model non-stationary time series. The ARIMA process
integrates the non-stationary data to transform into stationary data by applying
the difference operator. The difference operator is depicted as δ d , where d is the
number of different terms. For instance, the first-order difference operation can be
defined as δ 1 = xt − xt−1 . The graphical representation of an ARIMA(p, d, q) is
depicted in Fig. 2.4, where p, d, and q are the model terms or weights associated with
autoregression, difference, and moving average respectively. The workload values
at time t obtained after applying the difference are denoted as x̃ t . The first order
ARIMA is one of the simplest models that can be represented as shown in eq. 2.4,
where the term B denotes the backward shift operator (B × xt = xt−1 ) [91].
δd ... δd δd
φ1
ξt−q ... ξt−2 ξt−1 φq φ2 x̂t
θ2 θ1
θq
Prediction
fARIMA
ℵt
xt−1
fES
Forecast
(1 − α)
x̂t−1 x̂t
server workloads is shown in Figs. 2.6a and 2.7a using MAE and MASE respectively.
The model forecasts Calgary Trace with the least MAE for the prediction window of
length 5, 10, 20, and 40 minutes. For the prediction window of 60 minutes duration,
the least mean absolute error is obtained on the forecasts of Saskatchewan Trace.
Similarly, the MASE-based forecast results are depicted in Figs. 2.6a and 2.7b for web
and cloud server workloads respectively. It is evident that the D2 forecasts are most
accurate for the length of the prediction window of 5, 10, 30, and 60 minutes as per
the MASE. For 20 minutes of the duration, the best forecast accuracy measured using
MASE is generated for D1 . Based on the results, the first-order autoregressive process
models the web server workloads with better accuracy than cloud server workloads.
For cloud server-based workloads, the memory trace (D5 ) obtained better results over
CPU trace. In general, the first-order autoregressive process learns the pattern from
Calgary trace in a better way.
Reining up his horse in the centre of the narrow and dusty road, Rodrigo
cried:
'Begone, rash youth, lest I have you disarmed and scourged!' replied
Lozano, lowering his lance however, as he knew that he who barred the way
would not stand on trifles. 'We are five to one.'
At the head of three hundred mounted hidalgos, 'all wearing gold and
silken raiment, with perfumed gloves, and caps of gorgeous colours,' Don
Diego, now, as he thought, redeemed from disgrace, rode forth to meet the
King and kiss his hand, while Rodrigo repaired to the Convent of
Miraflores, with the blood-stained ribbon streaming from his casque, but
the face was not at the window now. Thrice he came thither and watched
and waited for it in vain, and believing that the Mother Abbess had
discovered his love-affair, he returned with a heavy heart to Burgos, to take
counsel of the King Sancho, though some say it was of this latter's father,
King Ferdinand.
But soon tidings came to the Court of Castile that a beautiful lady, who
had been foully wronged, was coming hither attended by a numerous train,
to seek justice at the hands of the King. All the young knights were ready to
embrace her cause, whatever it might be; but all, including the famous
Bellido Dolfos, withdrew in favour of Rodrigo, who first demanded to
make it his own; and yet he thought, 'God wot, why should I champion her,
when my own and only love is the Recluse of Miraflores?' And then the
sweet face at the window came before him in memory with all the soft
brightness of an opium-eater's dream.
Clad in black, with a gauze veil over her dark dishevelled tresses, her
eyes streaming with tears, the lady fell on her knees before the King,
exclaiming, as the Spanish ballad has it:
'Estrella mia!' he exclaimed, as she threw up her veil; 'can such sorrow
be? Are you Ximena Gomez?'
Spanish history makes the conquered kings five in number, and states
that he caused them to pay tribute after he set them at liberty, 'wherefore
they served him faithfully, and called him their Cid, or Lord.' It also records
that Ximena did not take the veil at Miraflores, but, curiously enough,
exhibited another strange sample of the manners of the age by petitioning
the King 'either to execute Rodrigo for killing her father, or give him to her
for a husband. The King chose the latter, and Rodrigo joyfully received
Ximena and took her to his mother, who kept her as her own child, and they
were betrothed; but Rodrigo promised to gain many more battles against the
Moors before he would claim her as his wife.' And so, while the Cid was
winning five provinces, and gaining glory too, with the edge of Tisona
among the infidels—of whom he slew an incredible number, till a saying of
his is a Spanish proverb to this day, 'The more Moors the more gain'—
Ximena spent her time in fear and hope among her favourite flowers and
love-birds at the house of Donna Teresa, in Burgos (Coronico de los Moros,
etc.).
And even after their marriage it was his boast, 'God wot! oftener is
Tisona than Ximena by my side.'
After the siege of Zamora, during which King Sancho was slain—
treacherously, it is averred, by Don Bellido Dolfos—the Cid, as the former
was repairing to Burgos, gave him a special message to Ximena:
'Tell her that I am coming; and, as an earnest thereof, give her this ring,
which I took from the hand of the Caliph of Cordova.'
Don Billido, who in his heart cherished a secret and treacherous love for
the betrothed of his friend, took the ring, and, saying emphatically,
'Rodrigo, amigo mio, haya cuenta sobre mi' (i.e., 'My friend, rely on me'),
rode gaily home to Burgos.
Bellido has been described as a man with a fierce hooked nose, a black
beard, and slightly treacherous eyes, that, if such are the true index of the
soul, might have revealed his natural character.
He gave the ring to Ximena, and told her that the Cid awaited her at
Miraflores. She was surprised at this, but, never doubting the comrade of
her intended husband, attended by two ladies, she set out for Miraflores,
closely veiled. They rode white palfreys, with velvet caparisons
embroidered with gold, and having silken bridles covered with little bells.
Bellido and some ruffians, on whom he could rely, formed their escort; but
they never reached Miraflores.
In due time the Cid Rodrigo came to Burgos with his heart full of
Ximena, his old love for her mingling with gratitude that she had forgiven
him for the terrible wrong he had done her, and already he seemed to see
her winning smile and her soft and lustrous eyes, that looked so truthfully
under the long, dark lashes that fringed them.
'At Miraflores, whither you sent for her,' was the reply.
'Or treachery,' said Donna Teresa; 'my mind misgives me, or I distrust
Don Bellido.'
'Can he have decoyed her away!' exclaimed the Cid, with alarm and
rage in his voice and eye.
But the old lady knew not what to think, and began to weep bitterly; and
still more did she weep when sure tidings came that in revenge for repelling
his addresses, the double traitor Bellido Dolfos had betrayed Ximena into
the hands of Hiaja, the savage Caliph of Toledo.
Rodrigo was beside himself with sorrow and dismay; but bethought him
at once of his sword, and prevailed upon his new master, Alphonso VI.,
King of Old Castile, to besiege the city of Toledo, offering him all his
knights for that enterprise.
The report of this siege, and the cause thereof—a Christian lady of rare
beauty and high rank, more than all, the betrothed of the Cid, being a
captive in the hands of the odious Hiaja—brought many knights and princes
from distant lands, particularly Raymond, Count of Toulouse, and two
princes of the royal blood of France, of the branch of Burgundy.
Their armies covered all the fertile plain amid which Toledo stands, on a
steep hill, round the base of which flows the Tagus. In some places the
spears of the infantry—whose massed columns seemed like a sea of
glittering steel—stood thick as upright corn; in others were the squadrons of
barbed horse, the knights and men-at-arms, all clothed in chain armour,
bright as winter frost or polished silver, their many-coloured plumes, their
square banners, and swallow-tailed pennons streaming out upon the wind.
High overall, with its towers and the minarets of its mosques, rose the
then infidel city of Toledo, the upper part of which was then, as now, girt by
Roman, and the lower part by Moorish walls. History tells us that when
Alphonso VI. had been a fugitive under the persecution of his brother and
predecessor, Sancho, he had found an asylum at the Court of the Caliph of
Toledo, who treated him with hospitality and princely distinction; and now
more than one Moorish warrior rode forth from the city to reproach
Alphonso with ingratitude to his benefactor, and many a terrible and
remarkable combat was fought under the walls of Toledo, among the
defenders of which was Don Bellido Dolfos, who had renounced his faith
and adopted the turban.
In the combats before the city, the Cid was daily occupied, and many a
Moorish warrior, horse and man, rolled in the dust beneath his lance or
battle-axe; and his followers were enriched by the spoil, the rare weapons,
the costly garments and jewels, that his hand won.
The hand-to-hand fighting was terrible, and the Christian knights, led by
the Cid, the Count of Toulouse, and others, dashed their horses through and
through the living tide of Moors that surged around them. Gorgeous as a
field of flowers, with their many-coloured turbans and flowing garments,
seemed the Moors as they kept shoulder to shoulder, guarding their heads
with round shields covered with glittering bosses, their sharp scimitars
flashing in the sun, their shouts rolling like thunder between the Tagus and
the walls of Toledo, as they fought with demoniac strength and ferocity, but
fought in vain. High over all the throng towered the Cid upon Babieca, its
mailed flanks stuck full of arrows and even broken lances.
'Santiágo y cerra España!' he shouted ever and anon—the old war-cry of
Spain—and he hewed on all sides with Tisona, till his sword-arm grew
weary, and the last who bit the dust beneath it was the traitor Don Bellido,
after whose fall the Moors were driven headlong into Toledo.
The siege lasted a year, during which Ximena and her two attendants
occupied a noble chamber in the palace of the Caliph. Its ceiling was
adorned with arabesques and fretwork, brilliant with gold and delicate
pencilling. In its centre was an alabaster fountain of perfumed water, and
round it were cages of gold and silver wire, full of singing birds; and there
daily the three ladies offered up their prayers on their knees for the success
of the Christian arms, and for their own release.
After a year and a day Toledo capitulated, and Ximena was restored to
the Cid, to whom all New Castile submitted, and who took possession of it
in the name of Alphonso VI.; and Madrid, then a small village, one day to
become the capital of Spain, was for the first time in the hands of the
Christians, and Hiaja was the last Caliph of Toledo.
To narrate all the heroic deeds performed by the Cid after his marriage
would require the space of a very large volume indeed. The great dominions
he acquired for his royal master the latter increased by espousing Zaid, a
daughter of the Moorish King of Andalusia, after which Rodrigo, at the
head of his knights, subdued the whole of Valentia. No sovereign prince in
Spain was more powerful than he; but he contented himself with the title of
Cid, and never assumed that of King, though he might easily have done so.
No warrior in Spain did more evil to the Moors, yet he occasionally joined
the Beni Huds of Zaragossa against the Counts of Barcelona, whom he
conquered twice. While he never failed in his word to a Christian, he
mercilessly despoiled the Jews, from two of whom he raised money for war,
by depositing with them two chests which were alleged to be full of plate,
but which contained only stones and sand.
Five years after the conquest of Valentia, worn out by incessant warfare,
he fell ill, and was abed when tidings were brought to him that Bucar, the
Moor, whom he had expelled from that kingdom, was advancing to regain it
with a mighty army of horse and foot; but Tisona lay idly in the scabbard
now. For seven days preceding his death, the Cid would taste nothing but a
little myrrh and balsam; and on the day he departed he took a solemn
farewell of Ximena, his kinsmen, and all his knights, whom he requested to
carefully bury his old war-horse Babieca, 'to the end that no dogs might eat
the flesh of him whose hoofs had trodden down so much dog's-flesh of the
Moors.' He bequeathed a coffer of silver to the two Jews, and desired that
his body should be borne to San Pedro de Cardena, and laid beside that of
his mother.
He died in the year 1097; but he who had been the terror of the Moors
for so many years when in life, was still fated to strike terror to them in
death, even while all the host of King Bucar were rejoicing that he had
passed away. At midnight, twelve days after that event, the Christians
prepared to abandon the city of Valentia—'Valentia of the Cid,' as it is
called to this day. His body, which had been placed, we are told, 'in a sitting
posture, and left to stiffen between two boards,' was placed on the back of
Babieca, upright in the saddle, with the feet tied in the stirrups. To all
appearance he was completely armed; a light shield of parchment, painted
with his device, was hung on his left arm; the terrible Tisona was fixed bare
and upright in his sword-hand. Geronymo, Bishop of Valentia, led Babieca
by the rein; Pedro Bermudez, with the banner of the Cid upraised, led the
van with 400 knights; then came the Cid's body, with Ximena and her
ladies, guarded by 600 men, and when day broke, though the Moors were
terrified to find that the Cid was there in his saddle again, a battle ensued,
and King Bucar was defeated; but Valentia was lost, and the sorrowing
warriors of Rodrigo continued their retreat to Old Castile and beyond the
Ebro.
At Olmedo they were met by his daughters, with all the knights of
Aragon, clad in black cloaks, with hoods rent, and their shields reversed at
their saddle bows; and with every religious and military solemnity incident
to the time, they laid him in his grave at San Pedro de Cardena, and two
years afterwards Gil Diaz, one of his most faithful followers, buried
Babieca before the gate of the church there. In the course of seven centuries
and a half the remains of the famous Cid Rodrigo have been removed
several times, the last occasion being by the French, in 1809, to the
Espolen, or public promenade of Burgos; but in 1826 they were restored to
San Pedro, where the tomb and effigies of himself and Ximena now remain
in a small but noble chapel. In that chapel lie the bones of Alvar Fanez
Minaya, whom he was wont to call his 'right arm;' of Pedro Bermudez,
Ordono, Martin Pelaez, the Asturian, and many more of his captains and
valiant friends.
His statue, as 'the dread and terror of the Moorish curs,' has a prominent
place in the quaint gateway of Santa Maria, erected by Charles V. at Burgos.
In the time of Cervantes the saddle of Babieca was preserved in the Royal
Armoury at Madrid, and Southey avers that he had personally seen and
handled Tisona, now an heirloom in the family of the Marquis de Falces.
On one side of the blade is graven, 'I am Tisona, made in the year 1002;' on
the other is the legend, 'AVE MARIA GRATIA PLENA DOMINUS
TUUM.'
THE BOY-GENERAL.
THE STORY OF JEAN CAVALIER.
THE BOY-GENERAL.
'Guillot—you here! Why have you left the mountain of St. Julian?'
'I have about as much beard as you, mon frère; and if my hand be
feeble, it has brought down many a wolf in Mialet and the Gevaudan,'
replied Guillot, slapping the butt of his carbine emphatically.
The speakers were young Guillot Cavalier and his elder brother Jean,
who was then, at the age of seventeen years, actually a general and second
in command of the Camisard army, the Insurgent Protestants of Languedoc;
who fought many a battle with Villars and De Montrevel, the best leaders of
the age; who, with Roland, led the great revolt in 1703; and who in his
twentieth year became a full colonel in the English army!
Both were very handsome lads, and both wore the white tunic (in
Languedocian, camisa) to distinguish themselves from their enemies, and
hence their well-known name of Camisards. Both were well armed, with
swords, silver-mounted pistols, and short carbines; but the elder wore over
his shoulder the scarf of a French general, and in his white velvet cap the
wing of an eagle. Strong—and tender as strong—was the bond of affection
between these two lads, who had both been born in the village of Ribaute,
among the pastoral mountains north of the Valley of Garden; and though
Jean was ready to face any peril and to 'do all that may become a man' for
the cause in which he had been so suddenly made a leader, and in which he
had already won such high distinction, his heart sank at the contemplation
of Guillot—a delicate boy, and their mother's chief care—encountering the
risks of that most savage and rancorous Civil War which now devastated
Languedoc.
Jean, as a very little boy, had been bred a shepherd, and was afterwards
apprenticed to a baker at Anduze; and it was from the employ of the latter
that, with a carbine in his hand, he went forth to become a Camisard, 'and
soon proved himself to be,' as history tells us, 'a most able general, as well
as a powerful prophet and preacher.'
'La Bonne Madelon is the mother we must serve just now, and I will not
quit your camp,' replied Guillot, whose eyes lit up, as he referred to one of
those wild, half-frenzied, and wholly enthusiastic prophetesses, or female
preachers, who thronged the camps of the Camisards, attended their
councils, and followed them into battle.
'Then be it so,' said Jean Cavalier resignedly; adding, 'I have good news
for you and all the faithful, Guillot. The Queen of Great Britain—the good
Queen Anne—is sending a fleet to our aid.'
In the 'Histoire des Pasteurs du Désert,' and other annals, we are told the
terrible story of that Civil War in which 30,000 Cevennois perished in battle
or on the scaffold, between November, 1702, and December, 1704. Well
fitted for desultory warfare are the mountains of Cevennes, with their rocky
labyrinth of deep gorges and dark defiles, which a mere handful of bold
peasantry were able to hold against the best troops of Louis XIV., and
where, to this hour, the population is almost entirely Protestant, inhabiting
some six hundred villages, which are all but inaccessible.
Roland and Cavalier levied their troops from the different parishes, each
of which furnished its quota of armed men and money, and fresh heroes to
fill up the vacancies in the ranks. Many believed themselves to be sword or
bullet proof, while 'the seizures, tortures, executions by breaking on the
wheel and burning alive (the common modes of punishing a Camisard), led
to reprisals on their part—to the slaying of priests and the sacking and
burning of Catholic churches.' But in the spirit of outrage, the French troops
were far surpassed by the guerilla bands, called Florentins, in the pay of the
Grand Monarque.
The sound of the drums had died away, but the sheeny bayonets
glistened in the sun, and the white Bourbon colours of the regiments, with
their golden fleur-de-lys, were waving in the wind, as the column of royal
troops began to penetrate a defile that was clothed with the olive, the vine,
and the fig-tree. The church and hamlet there had perished by fire; the place
was desolate; not a human being was visible, and without halting, the troops
pushed on, with an advanced guard to 'feel the way,' in front, till they
reached a portion of the defile where the impending rocks were higher, the
way narrower, and the trailing vines had given place to the dense, dark, and
woody luxuriance of forest trees. The flower of the column was composed
of one of the four battalions of the ancient regiment of Champagne, raised
so far back as the reign of Henry II.
'Halt!' cried the officer of the advanced guard, whose quick eye had
detected the bright flash of steel amid the green branches. In another
moment, a combination of fearful sounds burst like a storm upon the silent
air, while the soldiers halted, panting with the exertion of climbing the long
and steep ascent. An enormous fragment of rock, dislodged from above,
crashed with the sound of thunder into the defile below, a mass that must
have annihilated the entire advanced guard, had the officer not halted it in
time. Other masses of rock and rubbish came thundering down, barring all
advance, while more than a thousand voices made the defile re-echo with
the shouts of fierce exultation, mingled with a religious hymn.
On the fallen rock in front there was suddenly seen a female, 'the Good
Madelon,' kneeling in an attitude of frenzied supplication, her arms thrown
wildly up, her hands clasped, her black hair floating loose, her drapery
streaming on the wind, and by her side stood Cavalier. As yet no shots had
been fired.
'Voilà! 'Tis the rebel Cavalier!' cried De Montrevel, almost leaping in his
saddle with exultation; and his sharp words of command followed fast.
A volley was poured in front and on both flanks, and from these three
points it was closely responded to; and then the soldiers, who were in great
force, began, at the bayonet's point, to push up the woody sides of the
defile, firing as they went and driving the peasantry before them; and
meanwhile the prophetess—she of the supposed charmed life, La Bonne
Madelon, remained on her knees immovable, absorbed in prayer, half seen,
half hidden, amid the eddying smoke. Guillot strove to lead her aside, but in
vain; and when a bullet grazed his cheek, he rushed away to join his
brother, who, like him, strongly believed in the power of immunity from
death possessed by Madelon, and was now busy in the act of concentrating
and directing the operations of his scattered followers.
It is said that when the prophetess, whose eyes had in them the gleam of
insanity, felt the bullets whiz about her, a sense of danger came with the
sound, and that she opened her eyes and glanced about her, as if seeking to
escape, but she was grasped by four soldiers of the line; and that when the
Camisards beheld her feeble hands bound with cords, while her head sunk
on her breast, and she was dragged away, they became for a time panic-
stricken, and though they hovered on the precipice above the corpse-strewn
defile, they ceased to fire, and gazed on her conveyance to the rear in a
species of stupid wonder.
'She can save herself,' Cavalier is reported to have said, so perfect was
his belief, as a credulous mountaineer, in her divine mission; 'we cannot
rescue her now, but,' he added, lifting his cap and looking upward, 'some
miracle from heaven will.'
But no miracle was wrought, and with his solitary prisoner the Sieur de
Montrevel marched down, somewhat triumphantly, to the nearest town, the
white houses of which could be seen a league or two distant from the
mountains. That night Guillot, with a chosen party, stole from them, and
entered the silent street, from which all the inhabitants had fled, hoping to
find some trace of the Good Madelon, perhaps in the public prison, from
which they might see a way to free her.
But Montrevel and his men had departed, leaving in the market-place a
fearful object, which greeted the eyes of Guillot and his followers when
daybreak came in. Suspended by the neck from a gibbet in the centre of the
place hung the body of their prophetess in its well-known drapery, and
literally full of bullets, as the departing Florentins had made a target of it.
She had been a beautiful woman, whose husband and children had been
cruelly destroyed before her, and sorrow had doubtless turned her brain.
The fleet of Sir Cloudesley Shovel failed to land either succour or allies,
and returning to England, says Schomberg, in his 'Naval Chronology,' was
off the Isle of Wight on the 16th of November; so the Camisards now had
no hopes but in their own hearts and hands.
He had laid out his plans with wisdom, and alone, and a little apart from
his troops, was waiting the time to give them the signal to move, when from
all points around the Tour de Bellot burst forth a half-random storm of
musketry, and the boom of cannon announced that the King's troops were
upon him!
'By whom?'
And so it was; ere the Camisard outposts had been able to give the
alarm, they were cut to pieces, and only Cavalier and a few of his men were
able to sally from the tower before it was invested on all sides. Guillot and
others were shut up in it! Furious were the efforts made by Cavalier—
efforts urged by filial love and despair—to drive back the soldiers and
relieve those in the tower, from the windows and every cranny of which its
slender garrison poured a deadly fire for eight hours, till their ammunition
was expended, and then the edifice was set on fire; 290 perished in it, says
history, 100 Camisards lay dead outside, and around it were 1,200 of the
King's troops killed or wounded!
Compelled to retire some distance, yet fighting every inch of the way,
Cavalier beheld, with horror, the tower sheeted with fire. His soul died
within him as he thought of his brother, the boyish and gentle Guillot, and
all who were perishing there, and he strove to fight his way back just as day
was breaking, and by the light of it he could see, apart from all the hurly-
burly of the strife, a remarkable combat proceeding, and on the very verge
of a cliff close by.
That his brother Guillot might perish in battle, or by torture in the hands
of the enemy, Cavalier had always dreaded; but the catastrophe by which he
lost him was altogether unconceived: and the fortunes of the conflict led
him far from the vicinity of La Tour de Bellot, thus he could neither search
for the remains of Guillot, nor bestow funeral rites upon them.
For months the war went on. The bright valour and cool judgment of
Cavalier, 'the Boy-General,' for such he was, exalted him still more above
all other leaders of the Camisards, and especially so when he succeeded in
utterly defeating a considerable body of the royal troops at Martinarque,
under the Sieur de Montrevel, who commanded them.
The 6th of April, 1704, saw Cavalier again betrayed by one he trusted.
At the head of 900 foot and 300 horse, all well equipped, he entered the
Vaunage, or Valley of Noyes, so called from a little town of that name, in
the fertile district westward of Nismes, intending to waylay the Marechal de
Montrevel, who was on the way to Montpellier, but was himself lured into a
dreadful ambuscade, and surrounded on all sides by the royal troops,
including a great body of King James's-Irish, who had recently fought at the
battle of the Boyne.
On all sides burst forth from amid the shelter of trees and hedgerows the
withering fire of musketry, the boom of the cannon, and the hissing showers
of grape.
Eventually, however, his retreat was cut off, the royal troops occupied
every height, every avenue and pass that remained, and nothing was left for
him now but to cut his way out at all hazards, or die! He was not long in
choosing. 'Throwing aside his magnificent uniform and white plume, he put
on a common dress,' we are told, and ordering his comrades to close their
ranks, made a headlong dash at the enemy.
Cavalier, with the remainder of his force, escaped into the forest of
Cannes. This battle extended over all the ground from the mill of Langlode
to the town of Noyes. Of one thousand dead who lay on the field, one half
were Camisards. During the whole of the conflict one of their prophets,
named Daniel Gui, stood on the summit of a rock, amid six female
enthusiasts, three of whom were afterwards shot, invoking the God of
Battles to favour their cause.
But brief time had they for mutual explanations, as ere long the report
of musketry began to wake the echoes of the forest, and Daniel Gui came
rushing in with tidings that the Sieur de Lalande was putting to the sword
the entire inhabitants of the village of Euzet. Entering it suddenly, he had
found a bullock newly-skinned, and bales of hams, bread, and sausages
made up for the men of Cavalier, whom he at once traced and attacked with
vigour, and defeated with the loss of 170 men. Final vengeance now fell on
the unhappy villagers of Euzet, which, together with a cavern close by, was
found to be full of the wounded, ammunition, medicine, and stores of
Cavalier's forces. This sealed the fate of the former; and every human being
lying there was slaughtered, including the helpless creatures in the cavern.
Such was the awful system on which this war was carried on.
His presence there would seem to have been betrayed to the Duc de
Villars. At midnight, when he and his companions were fast asleep, the
sentinel on the tower-head suddenly heard amid the stillness of the hour the
distant noise of horses approaching at a furious gallop, and gave the alarm
just as a column of cavalry was entering the town.
On the 16th of August his body was dragged at the tail of a cart into
Nismes and burnt, while five of his companions were broken alive on the
wheel around his funeral pyre. Many Camisards perished thus here, the
most memorable executions being those of Catenat and Ravenel, who were
burned alive, almost within sight of the battle-field on which they had
defeated the Comte de Broglie.
Jean Cavalier found himself almost alone now, yet his spirit did not
quail.
Marshal Villars had now come to the conclusion that the warfare
seemed likely to become interminable; that it was possible to harass the
hardy mountaineers of the Cevennes, but not to conquer them. So resolute
was the spirit of the Camisards, so impregnable their hilly fortresses, that all
hope of ending the war so long as one was left alive, was relinquished by
this able officer; and we are told that in the heart of Cavalier, who beheld
the sufferings of the peasantry from incessant toil and famine, there rose a
great longing for peace, if it were possible with safety and honour; and on
ascertaining that 10,000 of the Huguenots were ready to lay down their
arms and submit to the king, he consented to hold an amicable parley with
any officer the latter might send.
Cavalier's first interview was with Lalande, who was sent by Marshal
Villars. 'Lalande surveyed the worn garments and pale cheeks of the young