Governance Policies in IT Service Support

Abhinay Puvvala and Veerendra K. Rai

Tata Consultancy Services, 54B, Hadapsar Industrial Estate, Pune- 411013, India

Keywords: IT Service Support, Agent based Model, Governance Policies, Upper Confidence Bound Algorithm.

Abstract: IT Service support provider, whether outsourced or kept in-house, has to abide by the Service Level

Agreements (SLA) that are derived from the business needs. Critical for IT Service support provider are the

human resources that are expected to resolve tickets. It is essential that the policies, which govern the

tickets’ movement amongst these resources, follow the business objectives such as service availability and

cost reduction. In this study, we propose an agent based model that represents an IT Service Support system.

A vital component in the model is the agent ‘Governor’, which makes policy decisions by reacting to

changes in the environment. The paper also studies the impact of various behavioural attributes of the

Governor on the service objectives.

1 INTRODUCTION

Communicating high level business objectives and

their relative importance from the IT Governance to

the IT Operations is a challenging task. The measure

of this challenge lies in understanding that Business-

IT alignment or the lack of it still remains one of the

major corporate concerns and the most critical

measure of the success of information technology as

a value adding component of business enterprises. It

is imperative that the design of each IT system is

aligned to business objectives without compromising

on the efficiency of IT systems (Sallé, 2004). In this

study, we focus our attention on the IT Service

support system. Specifically, this paper proposes a

framework for the governance of IT Service support

engagements.

Governance is a mechanism of course correction

when a project, program or engagement is in

execution mode to help projects meet the intended

outcomes. Scope of governance includes, among

other things, structural and organizational changes,

communications and policies. The scope of

governance in this study is limited to a set of rules

(policies) that includes assignment rules and pre-

emption rules in IT Service support engagement.

These rules play a key role in realizing the intended

objectives of the engagements.

Owing to the volatile nature of IT service support

engagements environments, these rules cannot be set

a priori and be expected to remain optimal

throughout the course of an engagement. Given the

constantly evolving business needs and their

possible repercussions on the IT systems, it is not

feasible to have a static set of rules. Another key

consideration while determining an optimal set of

rules is the interdependencies between them. We

propose an agent based game theoretic approach to

derive the optimal set of rules (assignment,

escalation and pre-emption rules) based on the

objectives and the context of a support engagement.

The rest of this paper is organized as follows:

Section 2 contains survey of relevant literature,

research gaps and the contribution of this study.

Section 3 describes the research model along with

discussions on relevant concepts from literature.

Section 4 contains the illustration of model proposed

in Section 3. Section 4 also has results of what-if and

sensitivity analyses. Section 5 concludes the paper.

2 LITERATURE REVIEW

Studies related to the governance of IT Service

support have primarily focussed on the following

research questions: 1. how to prioritize tickets based

on the business needs, thereby indirectly focusing on

the assignment rule, and 2. how to optimally divide

the staff amongst multiple shifts and technology

towers. While some studies (Gurvich et al, 2007;

Bassamboo et al, 2004) have attempted to address

these questions together, in most of research studies

512

Puvvala, A. and Rai, V.

Governance Policies in IT Service Support.

DOI: 10.5220/0006312805120519

In Proceedings of the 19th International Conference on Enterprise Information Systems (ICEIS 2017) - Volume 1, pages 512-519

ISBN: 978-989-758-247-9

these questions have been addressed independently.

Of these two questions, the focus of this study is

closer to the former rather than the latter. Bartolini

and Salle (2004) have proposed an approach to

present how business needs are used to prioritize

tickets and allocated to human resources.

In practice, basic and intuitive assignment

policies such as First Come First Serve (FCFS) and

priority based FCFS are often put to use. While

FCFS follows a strict first come first serve routine,

priority based FCFS gives precedence to requests

with higher priority. While these policies are

intuitive and easy to implement, they do not consider

the SLA norms, penalty costs etc. Assuncao et al

(2012) have studied the impact of both assignment

and pre-emption policies on ticket resolution and

service level agreement attainment. The dependence

of the policy optimum on the distribution of

workload is evident in their study. Lunardi et al

(2010) also have studied the management of changes

in the domain IT service support. Beyond that, there

is a vast amount of literature in the domain of

operations research on task scheduling (Rothkopf

1966; Pinedo 1995) in manufacturing that can be

drawn upon.

As mentioned in section 1, we use Agent Based

Modelling (ABM) to represent the engagement.

Agent-based modelling is an effective simulation

modelling technique that has grown rapidly in the

last few years. Agent based modelling is considered

a powerful paradigm to model human centric

systems like IT service support (Bonabeau, 2002).

The basic tenet of ABM is that a collection of

autonomous decision making agents that produce

emergent behaviour by interacting in an

environment under a given set of rules (Davidsson,

2002). This view resonates well with IT service

support system where various agents such as tickets

and resources who individually interact with each

other under defined processes which in turn are a

result of the policies. These interactions are analysed

by simulation of the agents’ behaviour. It is a

relatively new and emerging method in social

sciences, which can be applied to a problem by

defining a set of agents with related attributes,

behaviours and fitness function, the simulation

environment and the overall performance-measuring

objectives of the environment (Mataric, 1993).

A typical ABM model consists of an agent

having certain attributes, rules/actions, goals and

decisions to make. These defined agents are

generally governed by a fitness function. The aim of

creating a fitness function is that it allows multiple

agents of similar nature to have different attributes

by creating differences in parameters of fitness

function. This heterogeneity thus created is an

essential component of ABM and helps mimic the

real world more closely than other methods. These

countless interactions lead to ‘emergence’ of new

behaviour which had not been programmed into the

behaviour of the individual agents (Waldrop, 1992).

Agent based modelling has already been extensively

used in economics (Agent Based Computational

Economics (ACE)). Zaffar et al (2008) used it to

identify the impact of Variability of Open Source

Software (OSS) support costs, length of upgrade

cycle and interoperability costs on OSS diffusion.

Applications of Agent based modelling in IT

systems are limited. Jha et al (2014) have proposed

an agent based approach for estimating effort

required to resolve incidents in an IT support

engagement. In the next section, we discuss how

agent based modelling has been used to model IT

service support engagement.

Figure 1: Model Topology.

3 MODEL

3.1 Agents

Although the literature on Agent based modelling

does not provide a clear cut consensus on the

approach to identify agents in a system, there are

some basic guidelines that are common across

various definitions of agents. Bonabeau (2001)

considers any entity that has independent behaviour

governed by very basic reactive decision rules to a

complex and adaptive artificial intelligence. In

contrast, some researchers emphasize on the ability

of these entities to be adaptive to the environments

and have a learning component ingrained in their

behaviours. Casti (1997) separates these behavioural

elements into base level and higher level rules.

While the base level rules are meant to respond to

the environment, higher level rules can dynamically

change the base level rules (rules to change the

Governance Policies in IT Service Support

513

rules) by learning and adapting to the environment.

Jennings (2000) too emphasizes the need for

presence of agent attributes that are active rather

than purely passive. Active agent attributes are

essential for autonomous behaviour by which agent

can make independent decisions.

We adopt Bonebeau’s (2001) view to identify

agents in this system. The agent-set is a mixture of

autonomous, semi-autonomous and dependent

agents. Figure 1 has the topology of the agent based

model used in this study. Each agent is explained in

detail in the following sub sections.

3.1.1 Tickets

In IT production support, a ticket is an abstract unit

of work. Ticket can be any one of events, incidents,

problems, access requests and change requests.

Based on the business needs, tickets have to be

handled within specified time as directed by the

Service Level Agreements (SLAs). Typically, the

SLA terms are dependent on the ticket’s priority.

Priority is a composite of the urgency of the ticket

(how soon the business needs to be resolved) and

impact (how many users are affected by the ticket).

Also, tickets vary based on the type of skill required

for resolution. A “Technology Tower” signifies a

method of work organization usually employed in IT

production support where issues are grouped along

technical domains. Examples of technology tower

could be “.net”, “Java”, “SQL” etc. Naturally, the

skills needed to resolve tickets belonging to each of

these technology towers are different. While the type

of skill needed for resolution determines the ticket’s

technology tower, difference in level of skill needed

for resolution determines the level of support tickets

are routed to or eventually escalated to.

Since ticket handling is a knowledge intensive

task, a repository of all the information known about

tickets is maintained. The repository can take the

form of a set of standard operating procedures

(SOPs) or entries in Known Error Database

(KEDB). The effort needed to resolve tickets has

been observed to follow Power Law Distribution

(PLD). Based on the above described characteristics

the list of ticket attributes are shown in Table 1.

3.1.2 Resources

Despite the ongoing drive towards automation,

ITSM is majorly a human centric system. Tickets are

handled by resources, which are categorized into

multiple teams based on their skills and

specializations. In a typical IT production support

setup, tickets are responded and resolved by

resources. Response includes identifying, logging,

categorizing, prioritizing, routing and conducting

initial diagnosis of tickets. Whereas, Resolution

relatively is a more complex task. It involves

performing a set of steps to resolve a ticket. And, it

is done at the level of support that corresponds to the

ticket’s required resolution skills.

As given in Table 1 resources are characterized by a

set of static and dynamics attributes in our model.

While, technology tower, competency, cost,

likelihood of absence are the attributes that remain

static over the simulation. In contrast, ticket, shift,

net effort are the attributes that change dynamically

with the environment. The support structure support

in the model comprises of two technology towers

with teams divided into three and two levels of

support. Further, the support service is to be

provided 24x7, divided into 3 shifts of 8 hours each..

3.1.3 KEDB Agent

It is critical for any IT production support

engagement to record the knowledge acquired by

human resources in the process of handling tickets to

the extent possible. Of the multiple knowledge

management processes proposed in ITIL v3 (Cannon

et al, 2007), maintaining a well recorded Known

Error Database (KEDB) is vital to conduct efficient

IT service operations.

The purpose of a Known Error Database

(KEDB) is to store the knowledge of tickets– and

how they were overcome – to allow quicker

diagnosis and resolution when they recur (Cannon et

al, 2007). The first response to any service outage is

to quickly fix the issue and bring the system back up

to ensure service availability. The issue would then

be sent for root cause analysis, where a decision to

implement a change to prevent future occurrences of

incidents or update the KEDB with a workaround is

taken. The cost benefit analysis determines if there is

a business case for a permanent solution.

In our model, KEDB, as an agent, is

characterised by the following attributes. Integral to

the KEDB is its software efficiency which identifies

a new incident and matches with a KEDB record if it

exists. We codify the search efficiency of KEDB on

a scale of 0 to 1. The number of records/articles in

the database is the second attribute. The last attribute

is the overall efficiency of KEDB which directly

impacts the average resolution time. It is derived

from the other two attributes (number of articles and

search efficiency).

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

514

3.1.4 Governor

A key agent in our model is the Governor, who

makes policy decisions at the start of operations on

every day. In a real setting, this role is played by the

engagement manager. To replicate the cognitive

process of decision making by the manager, we

adopt Auer’s (2003) upper confidence bound

algorithm for exploration and exploitation.

Originally, designed for modelling the random

bandit problem (Robbins, 1952), the algorithm

models the problem of a gambler in a room with

multiple slot machines and has to decide which slot

machine he wants to play in each trail. It is

analogous to the problem of engagement manager

who has multiple policy options and has to decide

which one to adopt each day. As often is the case

with humans, while making policy decisions, the

algorithm assumes to have only limited knowledge

about the rewards associated with each policy

choice. Hence, occasionally the algorithm explores

various policy options to improve the knowledge

about rewards. Exploration, however, does not

necessarily improve the current payoff.

To account for the varying levels of exploratory

nature of the decision maker, a penalty term that dis-

incentivizes is added to the payoff. The penalty term

is multiplied by a quantifier that ranges between -1

to 1. The quantifier and penalty term are critical in

bringing the exploration and exploitation trade off

associated with making policy decisions. The

average of payoffs implies the current knowledge of

the decision maker and more importantly, facilitates

the learning aspect in the algorithm and also guides

agent’s future exploration. As the agent plays the

game more, i.e. gains more experience, his ability to

choose the optimal policy increases. Another key

aspect of decision making process is the ability to

respond to changes in the environment. By using a

sliding window that attaches more weightage to

newer policy runs, the agent accounts for changes in

the environment. A volatile environment mandates a

more responsive decision maker; hence a smaller

sliding window would be more beneficial. To start

with, each policy option is executed once during the

initialization period to compute payoffs. The payoff

(X) in our model is defined in equation 1.

∗∗













1



∗

(1)

F represents the penalty for each ticket that is not

SLA compliant. It is important to note that this

penalty is different from the penalty described in

previous paragraph. While penalty (F) is to choose

policies that minimize tickets out of SLA, the

penalty (P) described in the above paragraph is a

behavioural parameter of the decision maker.



SLA



represents the tickets resolved within SLA,





SLA



represents the total number of tickets that

missed SLA. λ is used to attach relative importance

between non compliant SLA tickets and total effort

(E) available for resolving tickets. Effort available

(E) is the product of number of resources and

number of hours each resource can work for and is

represented in person-hours. A policy that achieves

maximum SLA compliance while consuming

minimum effort is ought to have maximum payoff.

Expected reward of each policy option is computed

as shown in equation 2.













(2)

∗ln







⁄



(3)

While X



is the average payoff of the policy



over all the runs in the sliding window, P is the

penalty term that introduces the sensitivity to

exploratory nature of the agent while making

decisions. While B at -1 indicates extreme

exploitation, +1 indicates the exploration extreme.

Exploitation promotes use of policies that are tried,

tested and produced relatively better rewards.

Exploration strategy encourages the use of policies

that have not been used recently in search higher

rewards. B quantifies the exploratory behaviour or

risk taking nature of the policy maker.

Table 1: Agents and their Attributes.

Agent Attributes Agent Attributes

Static Dynamic Static Dynamic

Resource Technology tower Net Effort Governor Window Size Active Policy Set

Competency Ticket B Payoffs

Cost Shift Λ

P(absence) KEDB Search Efficiency Articles

Ticket Tower Net Effort Effectiveness

KEDB Entry Resource

SLA Shift

Competency

Priority

Governance Policies in IT Service Support

515

Table 2: SLA Violations and Average Effort times of Tickets.

Technology tower 1 Technology tower 2

Priority % Violations Average Effort % Violations Average Effort

Critical 2.97% 8.46% 28 min 4.59% 9.78% 17 min

High 41.47% 6.37% 146 min 38.97% 8.45% 197 min

Medium 40.60% 5.43% 3346 min 42.64% 6.66% 2876 min

Low 14.96% 3.86% 14547 min 13.80% 4.87% 16543 min

Some of the other key attributes of ‘Governor’ as

an agent are as follows. Window size is to fix the

number of periods that are considered for computing

average payoff. As discussed earlier, smaller

window sizes suit volatile engagements. Active

Policy is a dynamic attribute that changes based on

the prevailing set of governance policies. λ is used to

alter the relative importance attached between

effort/cost reduction and better SLA compliance

levels.

3.2 Policies

3.2.1 Assignment Policy

Assignment rules define how to assign tickets to

resources on the basis of priority, competency and

technology tower. Assignment policy decides the

order in which incoming tickets would be allocated

to a resource and to which particular resource they

are assigned to. The allocation of ticket to a resource

depends on various factors such as the type of ticket,

the expertise level required to resolve the, particular

competency required for resolution and whether

fungibility across levels and technology towers is

present. Fungibility here means resources are free to

move across levels and technology towers of support

to complete the pending tasks. A fungible structure

in production service engagements allows more

equitable distribution of work amongst resources

leading to higher utilization of resources and lower

waiting time for issues to be resolved. However, due

to reasons such as geographical distances, shift

timings, cost of resources etc. fungibility is not

always feasible.

3.2.2 Pre-emption Policy

Pre-emption rules outline the conditions under

which a resource can pre-empt the resolution process

of the ticket he is currently assigned to pick up

another ticket. Pre-emption policy decides whether

there would be an interruption to prioritize

resolution of some tickets over others in process at

any given time. Further, the interruption would be

based upon the priority of ticket or SLA time of the

ticket or both. Pre-emption policies also decide the

way in which overhead caused by pre-emption

should be handled.

4 EXPERIMENTAL RESULTS

To feed our simulation model, we used a ticket

workload log spanning one month. The total ticket

inflow during this period was about 1,839 tickets

spread over two supports technology towers of a

large financial services provider. The ticket log

comprises other relevant information such as arrival

times, priority, resolution time, effort time, time

spent at each support layer, SLA compliance and

reassignment reason. Some basic observations of the

ticket log are shown in table 2. In addition, the

staffing structure of the engagement is presented in

table 3.

To ensure the model conditions are reproduced

to the extent possible, the tickets are fed into the

model as it is from the ticket log. We avoided

deriving distributions from the log and regenerating

tickets within the model as the workload remained

constant in all the experiments. We evaluate

parameters such as SLA compliance, cost of

optimized resource set under multiple governor

configurations while observing the movement of

optimal governance policy set (Table 4).

Table 3: Staffing Structure.

Shift Technology tower Levels Resources

1 1 (1,2,3) (5,1,3)

2 (1,2,3) (3,3,1)

2 1 (1,2,3) (5,1,3)

2 (1,2,3) (3,2,1)

3 1 (1,2,3) (5,1,3)

2 (1,2,3) (2,3,1)

Cost $432645 SLA 95.36%

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

516

Table 4: Ticket Handling Policies.

# Assignment Policies # Pre-emption Policies

A1 No fungibility M1 No Pre-emption

A2 Fungibility across levels M2 Pre-emption based on Priority

A3 Fungibility across levels and technology towers M3 Pre-emption based on SLA expiry

Hybrid Policy Configurations

P1 P2 P3 P4 P5 P6 P7 P8 P9

A1,M1 A1,M2 A1,M3 A2,M1 A2,M2 A2,M3 A3,M1 A3,M2 A3,M3

4.1 Scenario Analysis

The purpose of this exercise is to feed the same

ticket log and see the performance under multiple

governor configurations. To start with, the three

governor configurations mentioned in table 5 are

used to see the policy movements. To evaluate these

configurations and their impact on SLA compliance

and effort reduction, a simulator based on the agent

based model discussed in section 3 has been

developed in Netlogo. On top of the simulator is an

optimizer that was built to produce the optimal

resource configuration given a workload, SLA

constraints and a set of governor’s policy choices.

Figures 2a, 2b and 2c show the prevailing policy

choice along with the payoff. The impact of changes

in the governor configuration is visibly evident in

the graphs.

Table 5: Governor Configurations.

Config 1 Config 2 Config 3

Penalties ($)

Low 6 10 6

Medium 8 12 7

High 15 14 9

Critical 20 15 11

Governor Parameters

λ 0.6 0.7 0.1

B -1.0 0.2 1.0

Window 7 days

Figure 2a: Policies-Payoffs for Configuration 1. Figure 2b: Policies-Payoffs for Configuration 2.

Figure 2c: Policies Payoffs for Configuration 3. Figure 2d: Policy changes across configurations.

Governance Policies in IT Service Support

517

The governor using the terms of reinforcement

learning at the start of operations on each day

represent the state of the environment and the choice

of an alternative represents the action of the learning

algorithm. The balance between these two

phenomena is shown by the difference in policy

choices of configuration 1 and 3, where the value of

B varies from one extreme to another. The policy

choices (Figure 2d) in each of these configurations

may not signify much standalone but when put

together with their corresponding SLA compliance

levels and cost parameters can provide interesting

insights. As shown in table 6, for the same context

(ticket workload, priority and SLA norms), changing

the Governor configurations can impact the

objective realization.

Table 6: SLA, Costs across configurations.

Scenario SLA Compliance Cost ($)

Config 1 95.98% 428617

Config 2 94.43% 427343

Config 3 96.87% 441667

λ signifies the relative weights attached to cost

and SLA compliance. Therefore, it is expected that

Configuration 3 to have more SLA Compliance than

Configurations 1 and 2. Similarly, Configuration 2

would focus make policy choices that minimize cost

considerations. In contrast, the implications of

parameter B, which signifies the exploratory

behaviour of the policy maker, are not so

straightforward. In the next section, we conduct

sensitivity analysis by varying B across the two

extremes of exploration and exploitation to

understand its impact on SLA compliance and cost

objectives.

4.2 Sensitivity Analysis

We divided the spectrum of B from -1 to 1 into a set

of 21 values spaced with a difference of .1. To

derive the relation between B and Cost, the

simulation is run for each of these values of B with

different resource configurations (number of

resources at each level, technology tower) before

zeroing in on the configuration that satisfies SLA

constraints with minimum cost. The optimizer that

was built to work on the results generated from the

simulator outputs the minimum cost.

The second part of the sensitivity analysis is to

derive the relation between B and SLA compliance.

To conduct this experiment, we have kept the

resources constant while varying the parameter B to

see the changes in SLA compliance.

Figure 3: Sensitivity Analysis (B vs Cost and SLA

Compliance).

The results are shown in Figure 3. It is

interesting to observe the magnitude of changes in

both black and grey curves despite the ticket

workload remaining constant throughout the

sensitivity analyses. Consequently, the impact of

governance policy choices on the goal realization is

very pronounced. In this case, with the given

distribution and frequency of ticket inflow, a value

close to .3 yields the best SLA compliance. In

comparison, a value of -.7 for B is better suited to

minimize the overall resource costs.

5 CONCLUSIONS

We have shown in this study the impact of

governance rules in an IT service support

engagement on the business level goals such as

service availability and cost reduction. The first

contribution of this study is to replicate the IT

service support system with an agent based model.

Central to the model is the Governor agent, which

plays the role of manager in an actual setting. Due

its parallels with the popular ‘Random Bandit’

problem, we have borrowed the Upper confidence

bound algorithm for exploration and exploitation

algorithm to model the cognitive process of the

Governor’s decision making. The agent is designed

to be autonomous and can independently make

policy choices based on the environmental variables.

The second contribution of this study is to connect

the Governor’s attributes/behaviour to key business

objectives such as SLA compliance and resource

costs. Since Governor’s attribute configurations

have a direct bearing on the policy choices, the link

between policy choices and business objectives is

incidental. An interesting direction for future

research is how to extrapolate the behavioural

attributes of the Governor to the manager of an

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

518

actual IT service support engagement, thereby,

establishing an association between the suitability of

a manager and the nature of support engagement.

REFERENCES

Assunçao, M. D., Cavalcante, V. F., de C Gatti, M. A.,

Netto, M. A., Pinhanez, C. S., & de Souza, C. R.

(2012, December). Scheduling with preemption for

incident management: when interrupting tasks is not

such a bad idea. In Proceedings of the Winter

Simulation Conference (p. 403).

Auer, P. (2002). Using confidence bounds for

exploitation-exploration trade-offs. Journal of

Machine Learning Research, 3(Nov), 397-422.

Bartolini, C., & Sallé, M. (2004, November). Business

driven prioritization of service incidents.

In International Workshop on Distributed Systems:

Operations and Management (pp. 64-75). Springer

Berlin Heidelberg.

Bassamboo, A., Harrison, J. M., & Zeevi, A. (2006).

Design and control of a large call center: Asymptotic

analysis of an LP-based method. Operations

Research, 54(3), 419-435.

Bonabeau, E. (2002). Agent-based modeling: Methods and

techniques for simulating human

systems. Proceedings of the National Academy of

Sciences, 99(suppl 3), 7280-7287.

Cannon, D., Wheeldon, D., Taylor, S., & Office of

Government Commerce. (2007). ITIL:[IT service

management practices; ITIL v3 core publications][4].

Service operation. TSO (The Stationery Office).

Casti, J. (1997) Would-Be Worlds: How Simulation Is

Changing the World of Science (Wiley, New York).

Davidsson, P. (2002). “Agent based social simulation: A

computer science view”. Journal of artificial societies

and social simulation, 5(1).

Gurvich, I., Armony, M., & Mandelbaum, A. (2008).

Service-level differentiation in call centers with fully

flexible servers. Management Science, 54(2), 279-294.

Jennings, N. R. (2000). On agent-based software

engineering, Artificial Intelligence, 117:277-296.

Jha, A. K., Puvvala, A., Mehta, S., Rai, V. K., & Vin, H.

M. (2014). An Agent Based Approach for Effort

Estimation in Production Support. Proceedings of the

25th Australasian Conference on Information Systems,

8th - 10th December, Auckland, New Zealand [179].

Lunardi, R. C., Andreis, F. G., da Costa Cordeiro, W. L.,

Wickboldt, J. A., Dalmazo, B. L., dos Santos, R. L., &

Bartolini, C. (2010, April). On strategies for planning

the assignment of human resources to it change

activities. In 2010 IEEE Network Operations and

Management Symposium- 2010 (pp. 248-255). IEEE.

Mataric, M. J. (1993). “Designing emergent behaviors:

From local interactions to collective intelligence”. In:

Proceedings of the Second International Conference

on Simulation of Adaptive Behavior (pp. 432-441).

Pinedo, M. Scheduling: theory, algorithms and systems,

Prentice-Hall, Englewood Cliffs, NJ 1995.

Robbins, H. (1985). Some aspects of the sequential design

of experiments. In Herbert Robbins Selected

Papers (pp. 169-177). Springer New York.

Rothkopf, M. H. (1966). Scheduling independent tasks on

parallel processors. Management Science, 12(5), 437-

447.

Sallé, M. (2004). IT Service Management and IT

Governance: review, comparative analysis and their

impact on utility computing. Hewlett-Packard

Company, 8-17.

Waldrop, M. M. (1992). Complexity: The Emerging

Science at the Edge of Order and Chaos. New York,

NY: Touchstone.

Zaffar, M. A., Kumar, R. L. and Zhao, K. (2008).

“Diffusion Dynamics of Open-Source Software in the

Presence of Upgrades: An Agent-Based

Computational Economics (ACE) Approach”, In:

Twenty Ninth International Conference on

Information Systems (ICIS 2008) Paris: Paper.55.

Governance Policies in IT Service Support

519