Table of Contents
Modelling Capacity Requirements for BBC IT Storage Area
Network: Experience and Research
Workneh Hailegiorgis! and Ddembe Williams?
Faculty of Business, Computing and Information Management
London South Bank University
103 Borough Road
London SE1 OAA, UK
1 e-mail: workneh.hailegiorgis@bbc.co.uk
2 Tel: +44 (020) 7815 7460 Fax: Tel: +44 (020) 7815 7499
e-mail: d.williams@sbu.ac.uk
Abstract
The research carried out in this paper provides BBC Technology Ltd, which is
the current IT services provider to the BBC with a system dynamics model as
a decision support tool in storage services management. BBC Technology Ltd
recently installed a SAN (Storage Area Network) which is a highly scalable
data storage infrastructure to satisfy an increasing level of demand by its IT
user base. The System Dynamics model developed in this study is used to
facilitate the understanding of SAN capacity utilisation trends and strategic
acquisition planning decisions. The model creates learning environment that
enables efficient and effective management of data storage capacity
requirements in different planning time frames. The results have significant
implications for long-term capacity investment decisions for IT Service
managers and capacity planning managers.
Key words: Storage Area Network (SAN), System Dynamics, Data Storage
Requirements, Modelling, IT Service Management, Decision Support Tool
1. Introduction
The BBC has over 34,000 IT Application users worldwide. It has one of the
largest IT infrastructure user base in Europe. This paper is based on the recent
acquisition by BBC Technology Ltd of the Storage Area Network (SAN)
infrastructure as a step change in storage strategy after experiencing year on
year exponential increase of storage utilization. As a media broadcasting
organization the demand for electronic data storage has been highly evident
as the corporation evolved to integrate the emerging digital broadcast
technologies which resulted in the introduction of media rich applications
generating large volumes of multi media data across the corporation.
Advances in computing technology have seen the corporate demand for
storage capacity requirements grow exponentially over the past decade and
this is expected to rise at an even steeper trajectory in the next few years
(Dataquest 2000). Disk storage requirement have been rising dramatically
enabled by the increasing speed of processing power, and the decreasing cost
of computer hardware and support. Business confidence in the electronic data
storage environment has increased as corporate data backup and retrieval
solutions have matured over this period, providing a high availability
platform for data sharing in rich media applications and e-commerce. Over
this period also, successive IT departments that provided the service and
support for the organisation’s information systems infrastructure have not
been able to address the problem fully or to provide a strategic solution to the
escalation of storage demand. In an attempt to contend with the problem in
the short term, different policies and practices have been implemented to
fulfil operational service requirements but none have attempted to fully
understand how data is generated, used and stored in the organisation, and
as a result, have not been able to provide a sustainable solution to deal with
the fast growing demand for storage.
After a number of years of escalating storage demand the management of
BBC Technology Ltd. recognised that the continuing growth of users’ data
required a step change in storage capacity planning and have embarked on
providing an advanced ‘state of the art’ solution to the capacity requirement
problem using Storage Area Networks (SAN) technology. As part of this
process, management have been forced to address the issues of storage
scalability, consolidation, centralisation, the administration cost and
availability of highly valuable data assets (Leman Brothers, 2001).It is
therefore important to fully analyse the strategic capacity requirements of this
new platform from both business and technical perspectives so as to enable
managers to accurately project and plan for future capacity expansions
resulting from data growth and associated demand for storage.
2. Information Technology Capacity Management Environment
2.1 Capacity management process overview
Storage Capacity Management is an unstructured activity. Kleijnen (1980)
points out that although the technical performance and storage utilisation
measurements using formula timing, simulation or other quantitative
techniques are highly structured, the economic evaluation of storage capacity
is an unstructured activity. In today’s environment predicting storage
utilisation and technological development over the next five years as the
system dynamics models attempts to do is a complex problem.
The IT Service Management forum using the ITIL ( IT Infrastructure Library)
framework defines Capacity Management essentially a balancing act that is
(Stewart, 1990) Cost against Capacity , the need to ensure that processing
Capacity that is purchased is not only cost justifiable in terms of business
needs, but also the need to make the most efficient use of those resources.
And, Supply against Demand, making sure that available supply of IT
infrastructure resources matches the demand made on it by the business, both
now and in the future, it may also be necessary to manage or influence the
demand for a particular resource.
(Stevens, 1980) indicates that the service objectives of timeliness, accuracy,
cost and reliability are important in the evaluation of IT storage capacity
management. He points out that the goal is not to achieve 100% utilisation,
but rather an acceptable level of service such that the IT infrastructure is
meeting the need of the users.
2.2 Current practices in capacity management
Capacity is usually one of the key considerations in defining strategic
objectives in Information Systems/Information Technology (IS/IT)
management. There are a number of mathematical models that are used to
define and resolve capacity planning problems. This paper examines models
used in addressing data storage capacity planning and forecasting issues in
the management decision-making process.
The BBC Storage Area Network infrastructure is managed by BBC
Technology Ltd, currently most of the capacity planning and management
functions are carried out through the manipulation of Microsoft Excel
Spreadsheet. Utilisation data from servers which are attached to the SAN are
collected on a fortnightly basis and a full summary is complied every month.
The collected data is used for primarily two reasons. The first reason is to
monitor and control storage utilisation and availability trends on the SAN
infrastructure and secondly to formulate capacity requirements plans based
on the trends observed in the operational environment. For the planning
process mangers apply formal estimation tools available in Microsoft Excel
such regression and time series analysis.
The in-built functions within Microsoft Excel are the primary tools currently
used in capacity planning process of BBC Technology Ltd. The application is
used to project values and create trend lines based on current storage
utilisation data recorded from the SAN file servers. To extend complex and
nonlinear utilisation data sets Microsoft Excel uses regression analysis, which
is a form of statistical analysis used for forecasting. (Microsoft Excel 2002)
Regression analysis estimates the relationship between variables so that a
given variable can be predicted from on or more other variables. The
worksheet functions used in capacity requirement calculations are forecast,
trend, and series functions.
The forecast function calculates, or predicts a future value by using existing
values. The predicted value is a y-value for a given x-value. The know values
are existing x-value and y-value, and the new value is predicted by using linear
regression. The known x-value is actual capacity utilisation and known y-value
is time interval. This function is used to predict future capacity requirements
based on current levels of utilisation.
The trend function returns values along a linear trend. The function fits a
straight line using the method of least squares to the arrays of know_y’s and
know_x’s. Returns the y-values along that line for the array of new_x’s that is
specified.
know_y’s : is the set of y-values we already know in the relationship y= mx +b
know_x’s : is the optional set of x-values that we may already know in the
relationship y=mxt+b
new_x’s: are new x-values for which we want trend to return corresponding y-
values
The series function is used by the capacity planner when manual control is
required to generate a linear or growth trend using the computer to fill the
projection values. In a linear series, the starting values are applied to the least-
squares algorithm (y=mx+tb) to generate the series. In growth series, starting
values are applied to the exponential curve algorithm ( y = b * m ‘x) to
generate the series. In either case the step values or the difference between the
first and next value in the series are ignored.
The storage management team in BBC Technology Ltd understand that the
spreadsheet functions applied on capacity utilisation data offers limited
modelling capabilities and require a lot of time to generate new scenarios and
therefore, decisions. They are looking for methods that generate optimal
decisions directly without the need for mangers to guess at optimal solutions
and analyse those decisions via spreadsheet manipulations. They also require
a system to respond quickly to changes and utilise less time for analysis.
3. Capacity Management Modelling with System Dynamics
The linear functions discussed using spreadsheet manipulations are part of
mathematical planning models that uses optimisation techniques to formulate
decisions in capacity planning. This type of decision-making is also referred
to as deterministic as it provides specific values given a set of conditions with
a set of objectives. It will input a set of parameters, which are pre-determined
over a planning horizon to output an optimised solution to the decision
maker. The assumption of “linearity” means that each causal factor impacts
the “effect” by a fixed, proportional magnitude. (Richmond 2001) Linear
functions could therefore be defined as a one way view of the problem domain”. The
analysis used in optimisation methods is static in nature with no feedback
factors that are causing changes into the system as long as the factors are
satisfying the set of objectives.
It is commonly recognised that the power of linear models is limited to
explaining past behaviour, or to predict future trends given that, there will be
no significant change in the pattern of behaviour that was observed.
(Daellenbach 1999) These models don’t fully understand outcomes which are
unintended, unpredicted and may partially or wholly negate the sought after
benefits of a particular decision.
Linear functions in spreadsheet manipulation have advanced with the
increase in computational power built on mathematical foundations and it is
used extensively in cost and price analysis in capacity planning and
management. However it has major short falls, in that it is not able to model
emergent behaviours, uncertainty and feedbacks without the need of
numerous assumptions, approximations and post model analysis. The
linearity conditions imply that non-linear effects such as economies of scale
and reliability issues could not be modelled accurately. Linear functions use
considerable computer resources to satisfy the large number of constraints
and variables in capacity planning. The complexity and size limits the
efficiency in providing an optimal solution. .
This paper purposes to apply System Dynamics modelling approach to solve
the capacity requirements planning problems identified in BBC storage
service management. In particular it will apply the Dynamic Synthesis
Methodology as the framework for analysis, model building and simulation
that will be discussed in this paper. (Williams 2002) Dynamic Synthesis
Methodology (DSM) refers to the integration of theoretical concepts and
structuring of parts and elements of a process over time in such a manner to
form a formal functional entity by synthesis as philosophy of science.
Synthesis is an attempt to fuse the findings of various branches of science into
coherent view, in order to explain why things operate the way they do.
Using DSM case study research method is combined with system dynamics
modelling to form a framework for aiding quantitative explanation and
prediction of the behaviour of complex systems under investigation. “The
advantage of modelling in system dynamics pseudo code is its iterative nature that
yields understanding which forms the basis for further analysis, theory testing and
extension” (Williams 2002). (Meadows 1982) The system dynamic modelling
approach enhances theory development using case study research method by
combining systems paradigms to natural science paradigms.
(Williams 2002) Case study research method is an empirical investigation that
probes and examines responses of convenient influences within the
operational environment of the task, users and system. System Dynamic
modelling approach in conjunction with case study research method will
provide the qualitative information that will be used to understand the
problem domain in more detail. As the DSM combines the case study and
simulation research methods it becomes a powerful tool for problem solving
and analysis. Case studies on their own are used to capture a description of
real situation while simulation experiments are used to build abstraction of
the real world and test the abstraction with formal data analysis. These
methods are complimentary which makes DSM a powerful research tool in
Operation Research and Management Science environment.
The DSM includes six iterative research process phases, namely: Problem
Statement, Field Studies, SD Model Building, Case Studies, Simulation
Experiments and Model Use and Theory Extension. The process diagram that
outlines this iterative research process, which is used to conduct the research
study, is illustrated on Figure 1 overleaf.
?
Problem Statem ent ae
Conduct a
Interviews
NewProblems and + sQbseneten
Issues discovered Field Study Stateteiat
/ Reference
é Modes historical ——
*s.._behaviours
2
System D ynamics
Model Building [7
Validate Model" +
eit ee Case Study
Test for Validity = ™ a Develop newmodel
4 Reliability ¥ or extend existing
Simulation theories incorporating
Experiments: new explanations
yo TestPolicies!
\. Theory Testing i Model Use and
fe Theory Extensions
Figure 1 The Dynamic Synthesis Methodology (Williams 2002 )
The usefulness of SD for developing and maintaining capacity requirements
process modelling and analysis lies in the following (William, Hall and
Kennedy, 2000; Williams, 2001)
¢ System Dynamics SAN Capacity Requirements decision making
process model permit the modelling of complex processes with various
levels of granularity; that enables the concentration of required level of
details without being overwhelmed by lower level of details.
¢ The SAN Capacity Requirements decision-making process model
developed can be used by stakeholders in training situation as one can
simulate the process to gain better understanding.
¢ The use of system thinking and application of SD modelling notation
gives the operational managers with a balanced perspective between
“hard” ( Davis and Vick 1977, Boehm 1981) and “soft” (Checkland and
Scholes 1990) systems problem solving paradigms.
IS/IT operational management are now more highly socio-technical processes
which need approaches or problem-solving paradigms that can captured both
the qualitative and quantitative issues commonly found in complex systems.
(William and Kennedy 1997) Paradigms that support both hard and soft
issues should be welcomed in operations management.
4.0 Dynamic Hypothesis of the Capacity Management Process
The BBC’s storage requirements have been growing at an exponential rate
over the past decade. As business requirements for desktop applications
started to increase, the data growth rate has been increasing at over 50% -70 %
on a year on year basis (EMC Final ITT Commercial Response, 2001). The
increase of applications used on the desktop has further intensified as the BBC
delivers its digital broadcasting output through the integration of PC based
machines running media rich applications for internet and broadcast services.
As a result of the ever-increasing demand by the business, BBCT had to make
a step change in introducing Storage Area Networks to satisfy the data
growth rate by increasing operating capacity using scalable and centralised
storage infrastructure.
As more data is generated and a proportion of total storage capacity becomes
operational, the available capacity for data storage goes down. As the rate of
growth in data stored on the BBCT storage infrastructure increases, the
available capacity provided is utilised at a faster rate and becomes operating
capacity. As available capacity decreases the operating capacity increases, the
live data storage environment BBCT needs to support and maintain. The
reduction of available capacity for storage means the reduction in data
growth rate and increase in service level violations as utilised operating
capacity exceeds the agreed service levels of availability. The increase in
service level violations is an indictor of the interruption of service continuity
and the reduction in data production that is vital to the organisation’s
business operations. It is the reduction in data production that has a slowing
effect on data growth rate until available capacity has been increased to a
level where the production of data can resume at the rate required by the
business. The relationship between data growth rate, available capacity,
operating capacity, service level violations and data production is a
relationship that has been driving the level of total storage capacity at an
exponential growth rate and this is the central problem that is under
investigation in this dynamic hypothesis.
4.1 Problem Statement
In the drive to meet business requirements and provide the agreed level of
service quality, BBCT is under constant pressure to balance the delivery of
cost effective IT solutions while maintaining an expected level of service
quality to its customer base. However, with a scaleable capacity expansion
technology now in place using EMC’s Storage Area Networks the cost of
capacity acquisition will also escalate dramatically if capacity requirements
are not planned effectively and demand for storage with resulting data
growth rate is not fully monitored and controlled. The characteristic of the
problem indicates a vicious circle of capacity requirements with increasing
levels of data storage and acquisition cost at every round of the planning
process. This balances the demand from the users to produce and store
exponentially increasing data assets. When this is viewed against BBCTs’
obligation underpinning its relationship with the BBC as its main customer
base, reducing IT Service costs looks difficult to achieve with escalating
storage capacity cost driven by increasing data growth rate and growth rate
projections. The challenge for BBCT is to provide quality storage service to
satisfy users’ storage capacity requirements and at the same time enable the
optimum utilisation of acquired operating capacity cost effectively. The
dynamics described in this hypothesis are used to define the relevant system
variables in the capacity requirements planning model.
4.2 Defining Key Variables
The problem statement defined above suggests that the system dynamics
Capacity Requirement Planning model could be characterised by the
fluctuations in users’ desktop application data production, available storage
capacity and service level violations resulting from the operating capacity
provided by BBCT to the customer base and the rate at which this storage
environment is utilised for data production. These important variables create
the framework for the quantitative planning process in relation to capacity
requirements. With the aid of simulation techniques it will be possible to
understand the effect of the changes in the variables over time and to make
the optimal decision that is able to balance cost against capacity and supply
against anticipated demand. The following key variables identified in the
problem statement and confirmed by BBC field observations in IT service
operations are important to explain and describe the changes in the capacity
requirements over different time frames. These key variables are as follows
{units of measure or dimensions are given in curly brackets};
BBCT Customers: The number of customers signed under the BBCT service
level agreement that have network storage requirements and are able to
generate and store data on BBCT SAN infrastructures {Users}
Desktop Application Data Production: Relates to the total amount of data
generated by BBCT customers connected to the server types hosted on the
SAN. These server types are currently Personal Data Servers (FS), Shared
Data Servers (RD) , E-mail/Public Folder Servers (XU). {Megabytes}
Data Growth Rate: Is the rate at which the data type stored on the SAN
increases over a given period of time. {Megabyte/ year}
Operating SAN Capacity: This is the amount of storage space that BBCT
customers are utilising, and have already populated with data types. This
includes capacity allocation through storage mirrors and resilience building
storage configurations. {Megabyte}
Available SAN Capacity: The amount of storage space that is provided under
the service level agreement for the utilisation of BBCT desktop system users,
also known as headroom capacity {Megabyte}
Total SAN Capacity: This relates to the total amount of SAN capacity that is
fully owned and managed by BBCT which is the total sum of Operating SAN
Capacity and Available SAN Capacity. {Megabyte}
Service Level Violations: is the total amount of storage capacity that is
utilised over and above the threshold of the agreed service level of storage
availability for the data types under investigation. Or it could represent the
quality of service provided by BBCT under capacity levels. {Megabytes}
Additional Capacity Acquired: relates to the total amount of storage capacity
acquisition derived from requirements of new businesses and expansion of
the operational environment in maintaining the agreed level of available
capacity. {Megabytes}
Capacity Required for New Business: The capacity that will be allocated to
new services that will be hosted on the SAN in consideration of available
capacity and data growth rate of the service. {Megabytes}
Operational Capacity Expansion Required: The capacity requirements
derived from operational data storage scarcity as indicated by service level
violations and deficiencies in quality of services. {Megabytes}
4.3 Reference Mode Behaviours
Reference modes can provide an insight into the dynamics that is present in a
system under research. The inherent problem facing SAN capacity
requirement planning is the exponential data growth rate with high
trajectory, the absence of policy to optimise the operating SAN capacity,
inability of the service provider to influence the demand for storage under
service level agreements and the escalating total cost of storage ownership.
The reference modes define the behaviour for the key system variables under
review and give an insight into their attributes, relationships and interaction
in an attempt to fully explore the problem area. The understanding gained
can be used to hypothesize possible effects of leverage applied on key
variables such as operating SAN capacity , data growth rate , desktop
application data production and additional capacity acquisitions. Figure 2
indicate the effect of data growth rate on available SAN capacity , applications
data production and service level violations as the number of BBC
Technology Ltd (BBCT) customer base expands over time.
Performance
Measures
Data Growth Rate
BBCT Customers.
4, 7 Desktop Applications
"Osta Production
é *
# Servite Level Violations
Available SAN Capacity
Time (Months)
Figure 2 Reference modes of Key Variable
10
Performance
Measures
Tetal SAN Capacity
Operating SAN
Capacity
Apastiona Capacity
x
oN sy toad
o* Operational Capacity
Exparsion Required
+
wet
fpacity Required for
+* New Business
Time (Months)
Figure 3 Reference modes of Key Variable
Figure 3 above indicates the effect of operating SAN capacity as BBCT
management continuously acquire more storage capacity to fix the
availability problems and attempt short term optimisation exercise on current
operating capacity. The net effect is an increasing total SAN capacity,
operational and new business capacity requirements and capacity
acquisitions. (Manni & Cavana 2000) The reference mode indicated by the
operating SAN capacity and service level violations can fit into three
identifiable system dynamics archetypes ‘fixes that fail’, ‘unintended
consequences of expediting’ or ‘shifting the burden’ as a result of a
continuous application of short term fixes only for the problem to reappear in
greater intensity which is characterised by the exponential increase of
additional capacity acquisitions and total capacity over time. The reference
modes indicated above provide a vital understanding into the underlying
dynamics present in the system.
The relationship between the key system variables in the SAN capacity
requirement model could be demonstrated using the causal loop diagram on
Figure 1.2 as a basis for constructing a dynamic decision support system. The
diagram provides a platform for research into capacity requirement planning
success and provides the key propositions that can be tested using data from
field studies and validated by stakeholders in SAN capacity. Figure 1.2
provides the theoretical framework of the underlying hypothesis proposed in
this paper. It is the relationship and interaction of variables as illustrated in
the diagram using cause and effect analysis that determine the behaviour
patterns over time as a result of effects from improvement in planning
capacity requirements, optimised capacity utilisation and cost effective
acquisitions. The initial capacity requirements planning feedback structure
11
contain eight dominant feedback loops of which two are Reinforcing loops (R)
and the other six are Balancing loops (B).
Data
Growth
+ Rate +
- RL +
Desktop Gf Operating SAN
Application Data AvailableSAN —-
Production Capacity _
+
f. X MaRS bd.
B3
BBCT Service
Level Total
Customers violations + Operational _ oohN
+ v3 - Capacity «= pacity
. p4)Expansion +
Capacity Asi) Required (BoA
Required for{ B5
New Business
4 Additional
Capacity
Acquired
Figure 4 Dynamic Hypothesis of the SAN Capacity Requirement
The interactions amongst the variables identified in the diagram determine
the capacity requirements levels on SAN. As users’ desktop data production
increases, the data growth rate levels accelerate, which reduces the available
SAN capacity at a much faster rate than anticipated, affecting service levels
and quality which will result in further capacity acquisitions. These increase
total SAN capacity, hence headroom available capacity and the cost to BBCT.
As the available capacity gets utilised the operating capacity also increases
which requires support and maintenance, increasing the total cost of
ownership to BBCT. As operating SAN capacity increase data growth also
increases, requiring further increase in availability. These continuous
dynamics for capacity requirements are explicitly illustrated in Figure 5.3 as
data growth rate -capacity availability loops (R1, B2 and B3): customer
storage demand - capacity supply loops (R2, B1, and B6): capacity acquisition
- service quality (B4, B5).
5.0 The Conceptual Framework for model design
The conceptual framework design covers the main areas of the storage
capacity requirement process that has been observed in the field study. This
will include variables, values, factors in the relationship between customer
and service provider that will be vital to include in the model development.
12
5.1 Design and Modelling Issues
The model design is attempting to address the following key issues: -
1.
2,
The understanding of data storage capacity in the context of IT service
management and service quality improvement.
Introducing simulation as an alternative to current linear projection
and decision making methods that are not addressing the exponential
demand escalation and applying effective capacity management
measures.
Using system dynamics to create understanding of the feedbacks in
storage demand and understanding the effects of formulating policies,
leverages and decisions.
Integration of strategic IT infrastructure management with defined IT
service management processes.
Provide a forum for addressing storage management for a large
enterprise that uses data assets in mission critical environments.
Defining and investigating key variables in the IT infrastructure
management that is critical to efficient and optimised utilisation of data
storage capacity.
The model is divided into to conceptual areas of business relationships. These
are the Customer/User Organisation and the Storage Service Providing
Organisation as illustrated in Figure 5 below.
13
Figure 5 Conceptual Frameworks of the SAN Capacity Requirement
Buninene
urement eesknent
tel |
|
—_ IT Service Performance
iatattion of
‘Gusires Reqlroner)
r ime, |g
1|- Rew sere operators serge
| tegu
Bathaae ae | —-renues Mow Series fo Storage a
irae stony Stowe ot Betng Srmone i
eee rs Provide IT Solution for the Business. TH + Development =
ane LS | capcom oem
Requirements I fo sane E opens
re
Model
|
[Operational Data Storeae
| ees eee
1 | exams one ae! ana Opera se ppc (ena ras naraaament |
etimiricroar — |____Stro Bin an cat ay Caecty anced [rsa |
(as renal Cheryl ieeecun, Lisi fih) Lan
jl ros ans seen ——$o, tran eng tec
\ restore / Mamata % reps «
\ X& Inrsuctr Renu gow a Salty SLA tree imi
ro cetera
“ neat
—_— a
| Unto Storag 9" "| Customer/User \! <> igs earret Enforce SLA. Storage
| Gigeteotion \! Organisations | “y" “sn
| |
pms ion
| || [Fee Fetcorcomeme [pastime Sa
I fe—stasezcoss chaos onsterageUse- f+ roftNaxinizaton “Proven [> Montor, analyse and tune
| Fea Stn sr Sens paren
ECT Curtoners, Tac coniceRewnun Save vel Tre! Level
| | —s.asneneon sions ob Tlaremecistrnge — [—"Ense >] artes
+ ears onrnentront | akege fatend ot somen
| il &_ sea
| \!
| er a
|
|
|
|
|
|
|
I
I
I
|
|
[1 |
Coat of Now |
I
|
|
I
I
|
|
I
The BBC desktop users are able to generate data storage requirements
through two different mechanisms. The first is through the daily operational
requirements through the core network services as defined in the model
personal, shared, e-mail data storage requirements. The second is through
strategic business requirements that are generated by the BBC business
strategy, the business plan and over all strategic business IS/IT requirements.
The Technology Direction Group (TDG) in the BBC controls this area.
The demand generated by the user/customer organisation affects different
stakeholders in the storage service-providing organisation. The model design
needs to address the different and at times conflicting interests of the
stakeholders. Table 1 below will summarise the various aspirations and
interests to different stakeholder in the storage capacity requirements
decision-making process.
14
Table 1 Main Stakeholders in the Capacity Requirements Model
Stakeholder Aspiration and Interest
Customer/User Require an IS/IT system that is reliable , effective, fast ,
easy to use and productive
Data Storage Infrastructure (SAN) Operational manage the SAN system and server
Management infrastructure against a specified SLA with capacity and
availability target as key performance indicator want to
meet SLA target and optimise storage
Data Storage Service Management Management of the SLA, monitoring the target are met
and analysing exceptions and SAL violations. Need to
maintain service quality in consistent fashion.
Business/IT Infrastructure Development of the storage services, capacity
Development acquisitions, server commissioning. Evaluation and
introduction of new technology. What to strategically
improve systems and services.
Finance Management What to reduce total cost of ownership , optimise retum
on investment and maximise revenue and profit
EMC Corporation What to provide quality service to BBCT, maximise
revenue and profit and maintain strategic business
partnership with BBC and avoidance of service penalties
The interaction between these competing and at times conflicting sets of
interest and aspirations between the various stakeholders form the bases by
which model elements are constructed and the process structures and
characteristics are defined in the model.
The field study was carried out during the months of March - October 2003.
The field study carried attempted to understand the Customer/User
organisation and impact of effective capacity management form the
perspective of service quality provision. Operational root-cause analysis
documents were examined in the field environment to determine persistent
capacity related problems stretching over a number of years. Focussed
interviews were used to determine the major drivers for the storage service-
providing organisation. The study needed to understand the critical success
factors and key performance indicators in determining capacity requirements
levels and identify the key variables that are used as inputs and outputs to the
model.
The model had both hard ‘quantitative’ and soft ‘qualitative’ data inputs that
were used to create the desired capacity requirement data outputs for the
planning of capacity requirement decision-making process. The ‘hard’ data
collection was centred on the customer/user management process, capacity
demand generation process, the SAN capacity supply management processes,
service delivery control process and the financial appraisal and management
processes. The ‘soft’ factors are of major importance in the capacity
requirement decision-making process that are not tangible and therefore are
difficult to quantify. In System Dynamics modelling and simulation it is
15
possible to incorporate soft variables with the use of graphical functions by
assigning variable values on scale of 0 -1.
6. Implementation of BBC SAN Capacity Requirement Model
System Dynamics (SD) methodology enables the modelling of data storage
utilisation and requirements planning processes. Using SD models it is
possible to identify the major system variables and their measures of
interaction within IT Service Management domain. It is possible to
understand the behaviour of users’ storage requirements and _ the
consequences of capacity planning decisions using, ‘what if?’ dynamic
analysis to optimise capacity acquisitions for IT service providing
organisation. The model could be used to identify optimised capacity levels
by balancing the total cost of ownership against capacity acquisitions and
supply of storage to users against demand made on the IT storage
infrastructure. (Williams 2002) SD models are rational structures that generate a
formal representation of behaviour of the system being studied. Using SD modelling
it will be possible to explore various approaches in providing the required
capacity to the business so that the agreed level of storage service is
maintained to users which heavily rely on the availability of storage capacity
at the right time, at a price acceptable and competitive in the IT Service
provision industry. The key question is how the effectiveness of storage
capacity requirements planning can be increased to improve the quality and
performance of IT service provision to the BBC, while reducing the cost of
capacity acquisitions by optimising capacity utilisation and accurate future
requirements projection.
The model development is based on the ITIL capacity management process
framework. ITIL provides a generic structure in IT service management that
can be used to identify the entity classes which enables the construction of
model segments and sectors. The fundamental building blocks are the
formulation of a relationship between the user/customer organisation and the
service providing organisation connected through a capacity demand and
supply processes
The model is divided into individual subsystems that are based on the
conceptual frame work defined previously in this paper and the key system
variables captured in the causal loop diagram formulated during the
development process.
6.1 Model Sectors
The model has four interacting subsystems and these are classified as SAN
Capacity Demand Management, SAN Capacity Supply Management, SAN
Capacity Service Management and SAN Capacity Finance Management. Their
associated 17 sectors are presented on Table 2. This represents the problem
16
statement and dynamic hypothesis and discussed earlier. The sector names
are based on the ITIL framework that identifies the core processes within
capacity management activity. Table 2 summarises the major function of each
sub system and their sectors represented in the model.
Table 2 Model Sectors and their Main Function in the SAN Capacity
Requirement Mod.
lel
Subsystem
Sector
Major Function
1. SAN Capacity Dem
and Management
1.1 User Management
1.2 Personal Data Storage Demand
1.3 E-mail Data Storage Demand
1.4 Shared Data Storage Demand
1.5 Data Archiving
The main function is to
monitor and quantify the
aggregate demand
generated on the SAN by
BBC desktop users data
production.
2. SAN Capacity Supply Management
2.1 Personal Data Servers Capacity Supply
2.2 E-mail Data Servers Capacity Supply
2.3. Shared Data Servers Capacity Supply
2.4 Operational SAN Capacity
Requirement
2.5 Strategic SAN Capacity Requirement
2.6 EMC SAN Capacity Reserve
The main function is to
control the supply of
storage capacity on the
SAN to satisfy demand.
Provide storage
requirement decisions
based on utilisation and
availability.
3. SAN Service Management
3.1 Service Level Management
3.2 Storage Service Performance
The main function is to
maintain the Service Level
Agreement Targets on the
SAN. Monitor, analyse
and tune service
performance. Manage the
demand of storage
services.
4. SAN Capacity Finance Management
4.1 SAN Capacity Cost
4.2 SAN Capacity Revenue
4.3 Excess Storage Charging
4.4 Service Penalties
The main function is to
manage the finances
associated with capacity
utilisation and acquisition.
With focus on total cost of
ownership, profit
maximization and retum
on investment.
The model sectors presented in Table 2 consist of approximately 238
equations in STELLA v8.0 and of this 23 are levels and 51 are rates. There are
17
17 interacting sectors contained in the four subsystems presented above. The
SAN capacity management process environment is divided into four major
sub-systems. The SAN Capacity Demand Management subsystem primary
objective is to monitor and quantify the load or storage requirements that will
be placed on the SAN infrastructure by the BBC user base. The SAN Capacity
Supply Management subsystem primary function is to supply capacity
required for SAN Capacity Demand Management subsystem by analysing
and projecting data growth rate across the server estate attached to the SAN
infrastructure. The SAN Service Management subsystem primary function is
to control the service delivery quality through key performance indicators
that are specified in the Service Level Agreement (SLA). The SAN Capacity
Finance Management subsystem is responsible for all finances management
activities associated with the utilisation and acquisition of SAN capacity.
6.2 Key Model Equations
The model has seventeen endogenous interacting sectors within the four
identified subsystem. Using STELLA software version 8.0, the variables
identified are used to characterise the capacity management process and their
relationships are explicitly defined and modelled for simulation experiments.
The model simulates for 15 years planning horizon with DT set at 0.25
indicating that monitoring activity summary is carried out on a quarterly
basis in a year period.
The equations for the variables New _Customer_Request_Rate and
Old_Customer Removal_Rate define the rate at which users accounts are added
to the system increasing the demand for storage resources and rate at which
users leave the organisation hence releasing storage resources. The interaction
of these equations defines the level of BBC Customers that will generate
storage capacity requirements on the SAN.
New_Customer_Request_Rate =
Normal(Average_New_Customer_Request_Rate,BBCT_Customers)/Customer_Request__Adjustmen
tTime {user/yr}
Old_Customer_Removal_Rate=
Average_Account_Deletions_per_month/Account_Deletion__Adjustment_Time {user/yr}
The type of data stored on the SAN can only be generated from three file
server types, which connect to the SAN forming the core network business
services for the BBC. These are personal data (FS) servers, e-mail (XU) servers,
and shared data (RD) servers. The three server types are incorporated in the
model to fully analyse storage demand put on the SAN. The demand
generated is characterised as data production and defined in the equations
below and these key equations determine the level of data storage
requirement in the demand management subsystem. The storage demand
generated is satisfied through the installed server infrastructure using the
18
SAN capacity supply management. The equations below represent the total
data production rate for the three server types that connect on the SAN.
Total_FS_Data__Production=
Average_No__of_Customers__per_FS_server*Number_of__FS_Servers*Data_Generated__per_cus
tomer {Megabyte}
Total_XU_Data__Production=
Average_No__of_Mailbox_per_XU_server*Number_of__Exchange__Servers*Mailbox_size_per_us
er {Megabyte}
Total_RD__Data_Prodution =
Average_No__Customers_per_RD_server*Number_of__RD_Server*RD_usage__per_Customer
{megabyte}
The three equations below characterise the data growth rate factors that are
observed in the field on the SAN infrastructure and are used to adjust the
required level of storage supply for the three server types connected to the
SAN.
FS_Data__Growth_Rate= CGROWTH(FS_Data_Growth__Rate_Factor)*FS_Server_Data
{megabyte}
XU_Data_Growth_Rate = CGROWTH(XU_Data_Growth__Rate_Factor)*XU_Server_Data
{megabyte}
RD_Data_Growth_Rate = CGROWTH(RD_Data_Growth__Rate_Factor)*RD_Server_Data
{Megabyte}
These growth factors for the different server types determine the level of
storage requirement generated by interacting with an example equation on
the personal data server defined below. The servers are operationally
monitored for SLA targets and operational requirements are generated based
on the availability thresholds defined in the SLA. This is calculated using the
following example equation on the personal data FS servers
FS_Data_Storage__Requierment = Actual__FS_Data_Storage__Availablity-
Target_FS__Data_Storage_Availablity {megabyte}
The capacity supply sectors operates using key understandings in operational
storage management dynamics that differentiates total capacity increase
through the expansion of existing servers or through the installation of new
servers. Both have the effect of increasing total usable storage capacity but
through different supply lines. These dynamics are captured in the following
equation.
Expanding_XU_Servers_Capacity =
(Total_XU__Servers_Capacity+XU_Server_Capacity_Expansion_Requierment)/XU_Server_Capacit
y_Adjust
Exchange_Servers_XU_Allocating =
(XU_Server__Mirroring+Storage__Requirement__per_XU_Server)*Number_of__XU_Servers/Exch
ange_Server_XU_Allocation__Time
19
There are differences between the processes of expanding the capacity on an
existing operational file server and allocating or installing new file server in
the SAN. When installing a new server additional capacity is needed for
mirroring, resilience and replication which are captured in the example
equations used for the E-mail Data Servers Allocating variable. The above
equations are used to supply all storage requirements generated from the
demand management sectors.
Operating SAN capacity stock is defined to provide storage capacity for all
the three server types requiring expansion due to the service level violations
resulting from high utilisation thresholds. This is expressed in the equation
below.
Operating_SAN_Capacity(t) = Operating SAN_Capacity(t- dt) +
(Supplying_Operating__SAN_Capacity - Expanding_FS_Servers_Capacity -
Expanding_RD_Servers_Capacity - Expanding _XU__Servers_Capacity) {Megabyte}
The available capacity stock defined in the equations below is the source of
capacity supply to new projects and allocation of new server capacity for all
the server types defined in the SAN. The two equations for Operating SAN
Capacity and Available SAN Capacity are at the centre of the SAN capacity
requirement model representing the relationship between the two most
important system variables of the model. These two variables are the key
decision making variables and together provide the model with the total SAN
capacity stock that can be managed by BBC Technology. These two variables
operate together to satisfy the storage demand requirements of the BBC.
Available_SAN__Capacity(t) = Available SAN__Capacity(t- dt) + (Expanding__SAN_Capacity -
Supplying__New_Projects_Capacity__Ca - Resource__Data_Servers_RD_Allocating -
Personal__Data_Servers_FS_Allocating - Supplying_Operating__SAN_Capacity -
Exchange__Servers XU__Allocating) * dt {megabyte}
The equation below Capacity requirements on SLA violations is a level that
represents the total utilisation of capacity above the SLA threshold. This stock
is an input to the rate of operating capacity supply
Capacity_Requirement__on_SLA__Violations(t)=Capacity_Requirement__on_SLA__Violations
(t- dt) + (Rate_of_Capacity_Requirement_on_SLA__Violation) * dt
INIT Capacity_Requirement__on_SLA__Violations = 0
The revenue generated in the model is expressed as a level using a rate
equation defined below. The rate is determined by the variables defined in
the equation Total Excess Storage Charges, SLA Revenue from SAN
Customers, SLA Revenue from the infrastructure and revenue generated from
project activities. The revenue from SAN customers are dependent upon the
number of users singed up on the Desktop SLA and SLA Storage Charges per
customer which are parameters observed in field study.
20
SAN_Revenue Rate =
(Total_Excess__Storage_ChargestSLA__Revenue_SAN_Customers+StorageInfrastructure_SLA_R
evenue+Storage_Revenue_on__New_Projects)/Revenue__Adjustment_Time {E/yr}
The equations defined below defines the Total Cost of Ownership of the SAN
infrastructure. Together with the storage lease cost defined in the model
determine the level of total SAN capacity cost.
SAN_Operating_Costs =
Human_Resource_Cost+SAN_Accomodation_Cost+SAN_Hardware__Maintenance_Cost+SAN_Sof
tware__Cost+SAN__Power_Consumption_Cost {£}
6.3 Simulation Results
The simulation results can be directly compared with the reference mode
behaviours identified in Figures 2 and 3. The level of operating SAN capacity
increases as the available SAN capacity decreases. The data growth rate is
expected to rise as the number of BBC users increase and this will increase the
level of capacity requirements.
Simulation results were generated using ten key variables that have been
defined in the dynamic hypothesis. Figure 7 and 8 show the simulation
outputs using the model. Figure 7 overleaf illustrates the relationship of
Operating SAN Capacity (1), Data Growth Rate [ represented by FS Server
Data (2), XU Server Data (3), RD Server Data (4)] ,BBC Customers (5). The
output from the simulation reproduces the base case behaviours in the
dynamic hypothesis.
9 1: operating SAN Cap... 2:FS Server Data 3: XU Server Data 4:RD Server Data ‘5: BBCT Customers
1 9e+009)
2 204012.
3 700000000}
4 40000000
5 2000000
4,5e+009
Hy
0.00 3.15 7.50 11.25 15.00
Page 17 Years 19:34 14 May 2004
qd eee 2? Base Case Behaviour 1
Figure 7: Simulation Outputs BBC Customer, Data Growth Rate
and Operating Capacity
21
Figure 8 presents the simulation output that shows the behaviour of key
variable Capacity Requirement on SLA Violations (1), Available SAN
Capacity (2), Total FS Server Capacity (3), and Total Exchange Capacity (4)
and Capacity for New Projects (5). It is possible to see a behaviour defined by
one of the key propositions in the dynamic hypothesis that when Capacity
Requirements (1) increases Available SAN Capacity (2) will being to decrease
9 i-capacity Requirem... 2: Available SAN Ca... 3:Total FS Servers... 4: Total Exchange Se...5: Capacity for New
1 6e+011
2 5e+010.
i esol
e
‘ ae ae |
| ao Peace
W
\
fe
0.00 3.75 7.50 11.25 15.00
Page 18 Years 19:34 14 May 200¢
q aee 2 Base Case Behaviour 2
Figure 8: Simulation Outputs Capacity Requirements, Available
Capacity and Total SAN Capacity
6.4 BBC Case Analysis and Model Credibility
The case analysis attempts to show the SAN Capacity Requirements Model is
capable of generating plausible data that is consistent with the reference
modes and dynamic hypothesis defined in this paper. The analysis correlates
data that is actually generated in the field of operation at the BBC with data
generated through model simulation. It assesses the model behaviour against
the operational environment to gain insight of the accuracy and credibility of
the implemented model.
The data available in BBC Technology that was recording utilisation trends
and data growth rate behaviours was available on Personal Data (FS) servers.
The data is automatically gathered from all FS servers in London where seven
FS servers are on the SAN infrastructure. The FS servers installed in the SAN
have a total capacity of 150 Gigabyte (150,000 Megabytes) and the servers
outside the SAN infrastructure (i.e. stand alone servers) are installed with
200GB or 100GB. On a fortnightly basis the servers’ utilisation is measured
against total server capacity and the figure is converted into % utilisation
22
against the total server capacity. These figures are recorded in Excel and a
trend analysis is carried out using a Time Series Analysis. The diagram on
Figure 9 presents the actual time series analysis done on sixteen FS servers in
London showing the utilisation over time behaviour. The x-axis presents the
time in months and y-axis presents the % utilisation.
FS Servers Operating Capacity
120.00
—=BBCFS2031
—secrs2024
——pecrs2018
——BecFs2028
I—ecrs2029
—BcFs2025
100.00
80.00 ——secrs2027
I—secrs2026
\—secrs2042
—secrs2002
60.00 | nace 2008
——secrs2009
—secrs2023
|—secrs2014
40.00 I———BBCFS2055
oo atcesizosz
20.00
0.00
29/08/2003 14/09/2003 28/09/2003 12/10/2003 26/10/2003 09/11/2003 23/11/2003 07/12/2003 21/12/2003 04/01/2004
Figure 9 Actual FS Server Operating Capacity at the BBC
BD és senerStroage:1-2-3-4-5-) 7-8-9-10
i 3000000™
\
1 20000004 z=
ct 1000000:
0.00 0.20 0.40 0.60 0.80 1.00
Page 26 Years 10:49 PM 20 ul 2004
N 3 BaF 9 FS Server Storage
Figure 10: Simulation output for FS Servers Capacity and Storage
Utilisation
Figure 10 presents comparative simulation outputs of the SAN Capacity
Requirements Model on FS Server Storage over a one year period.
23
The graph on Figure 9 shows the behaviour of live operational servers using
data gathered at the BBC and converted in a time series analysis with linear
approximation for a period a year. The graph on Figure 10 captures the
behaviour of seven SAN attached FS servers using the system dynamics
simulation model over a period of a year. It is possible to see the similarity in
behaviour as both graphs show an increase in capacity utilisation with
exponential characteristics. The similarity of the simulation result and the
actual case analysis indicates that SAN Capacity Requirements model is able
to capture real world business systems behaviours observed by the model
through the dynamic hypothesis. This finding confirms the suggestions on the
usefulness of system dynamics in capturing the intricate dynamics based on
feedback theory of the capacity requirements process.
7. Conclusions
The research described in this paper models the SAN capacity requirements
planning and management dynamics from a feedback perspective. The
success of balancing the demand and supply of data storage capacity at the
right price, while optimising the utilisation of installed operational capacity is
critical to the competitiveness of the IT services providing organisation. Its
competitiveness can increase by reducing the total cost of IT storage service
provision and increasing the level of service quality it provides to its
customers. The efficient and effective delivery of data storage capacity is
crucial to the performance of organisations that strategically depend on the
supply and availability of data storage capacity. The proliferation of
information systems and information technology as the backbones of business
operation make data storage capacity management a critical success factor in
strategic business management.
In response to these critical business requirements, IT service providers have
deployed high cost infrastructure monitoring tools to capture the utilisation
trends of the data storage systems and provide data analysis and tuning
recommendations in the operational environment. However the tools have
not been able to fully understand or capture the underlying business data
storage requirements or data production capabilities of the customer
organisation. Therefore the tools are only able to provide reactive capacity
planning functionalities based on monitored historical data using linear
projection techniques. The tools are not able to account for the feedbacks from
business requirements and associated demand side variables. The linear
demand projection methods have shown severe weaknesses applied on
exponentially volatile behaviours in data storage requirements. In this paper
the SAN Capacity Requirements Model applies the system dynamics
approach using the Dynamic Synthesis Methodology. It captures the
feedbacks generated with changing business requirements and associated
data production rates to formulate a clearer view of strategic storage capacity
requirements.
24
The research found that the level of data production combined with the
number of users accessing the storage infrastructure, the type and number of
desktop applications accessed by users highly impacted the levels of
operating capacity and available capacity. This indicates that analysing just
the level of data growth rate is not enough to make an accurate analysis of
capacity requirements. The experiments conducted on the system dynamics
model shows the level of operating SAN capacity is not sufficient to quantify
the level of business requirements for data storage. The level of operating
SAN capacity is dependent upon the finite level of available SAN capacity,
the sum of these two variable represent the total SAN capacity installed on
the storage infrastructure. This cause and effect relationship between
Operating and Available SAN capacity often excludes the users underlying
root cause dynamics of business requirements and data production
capabilities resulting in flawed data growth projections. The most likely
outcome is inadequate or excessive capacity acquisitions. The model is able to
generate various outputs for data growth rates and levels of SAN capacity
based on storage demand and data production rates, which are the key
variables in data storage requirement analysis. This is the first time this link
has been analysed, quantified and simulated in the organisation.
The understanding of IT capacity management and the importance of
business requirements analysis has been highlighted in structured IT service
management approaches. The ITIL framework defines modelling as one of
the key capacity management activities. This paper reinforces the application
of system dynamics modelling as a basis of strategic capacity requirements
decision making by understanding the feedback loops present in the
capacity requirements process. It is this feedback loops that research and
practicing managers can exploit by aligning operational IT knowledge,
technological innovation and organisational processes to reduce requirement
uncertainty and effectively optimise capacity acquisition and utilisation.
The SAN Capacity Requirements model could be the basis for ITIL based
capacity management modelling approach to provide a__ holistic
understanding of the business data storage requirements dynamics. The
model can be used to investigate the effects of volatile and unpredictable
changes in data storage requirements resulting from rapid changes in
information systems development and technological innovation.
25
References
Boehm, B., (1981), Software Engineering Economics, Prentice Hall, Englewood,
Cliff NJ
Burd, S.D. and Kassicieh S.K. (1991). ‘Decision Support for Supercomputer
Acquisition’ , Journal of Operations Research, Vol. 39, No 3, May-June
Checkland, P., and Scholes, J. (1990). “ Software System Methodology in
Action”. Wiley, Chichester.
Daellenbach, Hans G.,(1999). Systems and Decision Making - A Management
Science Approach, John Wiley & Sons
Dataquest 2000, Gartner Business Information, www.gartner.com access at
20.09.03
Davis, C. and Vick, C R. (1997), ‘The software Development System’. IEEE Trans
Software Engineering
EMC Corporation, (2002), BBCT SAN Final ITT Commercial Response
Kleijnen, J. P.C. (1980). Computers and Profit: Quantifying Financial Benefits
of Information. Addison Wesley, Reading Mass., 28-29
Ku A. (1995). Modelling Uncertainty in Electricity Capacity Planning - Phd.
Thesis for London Business School, University of London.
Leman Brothers. (2001), ‘Enterprise Storage Takes Centre Stage, What will be the
Encore?’ Leman Brothers.
Maani Kambiz & Casvana, R. Y., (2000). Systems Thinking and Modelling,
Prentice-Hall New Jersey
Meadows, D. M. (1982). Lessons from Global Modelling and Modellers
Futures Vol:14 pp:111-121
Microsoft Excel Corporation, Microsoft Excel 2002 Help, www.micrsoft.com
accessed at 20.06.04
Pidd, M (1992). Computer Simulation in Management Science. John Wiley
and Sons Ltd,UK.
26
Richmond, B. (2001), An Introduction to System Thinking, STELLA, High
Performances Systems Inc, Lebanon, NH.
Senge, P.M. (1990), The Fifth Discipline: The Art and Practice of the
Learning Organisation. Doubleday, New York.
Sherwood, D. (2002). Seeing the Forest for the Trees- A Manager's Guide to
Applying Systems Thinking, Nicholas Brealey Publishing, London.
Stevens, B. A. 1980. Audit and Control of Performance in Data Processing.
EDP Perform. Rev. 8, 1-13
Stewart. (1990). Capacity Planning, Central Computer and
Telecommunications Agency (CCTA), John Wiley & Sons, Inc, New York
Williams, D. (2001). A System Dynamics Theory of Requirements Engineering
Process. In proceedings of the International System Dynamics Conference
2001, Atlanta, USA. 23 -30July 2001
Williams, D. (2002) ‘Dynamic Synthesis Methodology: A theoretical framework for
decision process modelling and analysis’. In the proceedings of the 1st European
Conference on Research Methods for Business and Management Studies, (Ed)
Dan Remenyi, MCIL, Reading 29-30.
Williams, D., (2003) ‘ Challenges of System Dynamics to Deliver Requirements
Engineering Projects: Faster, Better and Cheaper’. In proceedings of the 21st
International System Dynamics Conference, New York, USA, 20-24 July 2003,
System Dynamics Society
Williams, D., Hall, T., Kennedy, M., (1999). ‘A Framework for Improving the
Requirement Engineering Process modelling’. In Eds C. Hawkins, G. King, M.
Ross, G. Staples, Managing Quality: Proc of Software Quality Management
Conference Southampton UK 29-31 March.
Wolstenholme E. F. (1990) “System Inquiry - A System Dynamic
Approach”, Wiley, Chichester
27
Back to the Top
CC BY-NC-SA 4.0