Sie sind auf Seite 1von 12

Journal of Intelligent & Fuzzy Systems 36 (2019) 2311–2322 2311

DOI:10.3233/JIFS-169942
IOS Press

Adaptive neuro-fuzzy behavioral learning


strategy for effective decision making
in the fuzzy-based cloud service
negotiation framework
Rajkumar Rajavela,∗ , Kanagachidambaresan Iyera , R. Maheswarb , P. Jayarajanb
and R. Udaiyakumarb
a Department of CSE, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, India
b Department of ECE, Sri Krishna College of Technology, India

Abstract. Future cloud computing creates a new trend of opting service over the internet through some intelligent third-party
broker. In cloud market, both consumer and provider compete with each other against the conflicting requirements, and the
competition among cloud providers to trade their services to potential consumers of cloud market. There is an increasing need
for automated negotiation framework to quickly reach agreement in competitive cloud market which can provide maximum
utility value and success rate among the negotiating parties. Researchers develop various behavioral learning negotiation
strategies (such as market driven) in the existing negotiation frameworks for maximizing either the choice of utility value or
success rate of parties. Moreover these strategies can be applicable to the environment, where the opponent’s behaviors are
predictable or precisely known. It may be daunting to apply in the dynamically varying competitive cloud market. So, the
proposed Adaptive Neuro-Fuzzy Behavioral Learning (ANFBL) strategy can be applicable, where the opponent’s behavior is
partially and imprecisely known. Therefore, the proposed strategy can maximize both utility value and success rate without
compromising either choice. An extensive simulation is conducted to evaluate the efficiency of strategies which shows that
proposed strategy achieve higher utility and higher success rate than existing learning approach, without any negotiation
conflict among the parties.

Keywords: Cloud computing, automated negotiation framework, fuzzy behavioral learning, adaptive neuro-fuzzy

1. Introduction consumers, which can brings the similar comfort


of traditional utilities like electricity, water, and
In this decade, cloud computing is being used gas [2]. In recent years, cloud computing technol-
in different application domains such as indus- ogy has become an essential part of the enterprise
try, government, and science [1]. The major vision information technology business for its on-demand
of cloud computing is to offers huge computing self service provisioning and metering of services.
power and information technology service delivery to It offers high computing resource availability and
throughput for the services after confirming the SLA
∗ Corresponding author. Rajkumar Rajavel, Department of agreement between the participants (consumer and
CSE, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Sci- service provider) [3]. The SLA is a contract made
ence and Technology, India. E-mail: rajkumarprt@gmail.com. between the participants to promise the Quality of

ISSN 1064-1246/19/$35.00 © 2019 – IOS Press and the authors. All rights reserved
2312 R. Rajavel et al. / ANFBL strategy for effective decision making

Service as goals for the vision of cloud comput- This cloud-based resource provisioning will
ing [4]. The Quality of Service goals and economic dynamically configure the virtual machine, which is
conditions such as pricing and violation terms to be in need of an SLA negotiation between the service
established are clearly stated in the SLA. This SLA is consumer and service provider, using an intermedi-
classified into static (provider predefined) SLA and ate third party broker. The cloud agency can act as a
dynamic (negotiated) SLA [5]. In the former case, a broker for maintaining dynamic provisioning, mon-
generic model of the SLA template is provided to all itoring, and reconfiguration of the cloud resources
the consumers. In the latter case, the consumer and on behalf of the user [13]. Since, the cloud service
provider undergo a series of negotiation processes negotiation participants such as service consumer,
between them to accomplish a mutually agreed SLA broker, and service provider are independent bodies
template. with different requirements, policies, and objec-
In general, cloud providers like Amazon EC2, tives; there is a need for a negotiation framework
and Microsoft Azure define a general SLA doc- among the participants to resolve their differences
ument for all consumers that assure to guarantee [14]. In a real-time e-commerce negotiation problem,
99.9% service availability. This type of static SLA is the broker-based negotiation framework using other
automatically recognized as soon as consumer con- computing paradigms, like the grid and cluster, was
firms the service request through online credit card restricted to resource constraints due to its negotia-
payment. The provisioning of services through the tion complexity [15, 16]. Therefore, to overcome the
static SLA is a semi-customized service provisioning resource constraint, a negotiation framework needs to
mechanism which is in-compliance due to prede- be presented over the cloud environment, due to its
fined non-functional properties of Quality of Service elastic scaling of resources. Negotiation framework
goals specified by the providers. Current cloud man- is the composition of various automated compo-
agement systems provide a traditional SLA-based nents like negotiation strategy, negotiation protocol,
service provisioning and metering mechanism [6, automated agent based communication among the
7]. SLA refers to an agreement containing set of negotiating parties as defined in the next section.
functional and non-functional properties of the ser-
vice mutually agreed between the service consumer
and the service provider [8]. Only semi-customized 2. Introduction
service provisioning with static SLA is available in
today’s cloud management system. Forecasts indicate In this decade, cloud computing is being used in
that a consumer with specialized Quality of Service different application domains such as industry, gov-
requirements may not be satisfied by the provider for ernment, and science [1].
maximizing their revenue. In addition, the provider
cannot provide differentiated service provisioning to 2.1. Negotiation framework
consumers with specialized quality expectations.
To satisfy such future demands, an SLA-oriented Initially, the negotiation concept was used for
cloud management system mechanism is proposed reserving bandwidth in the SLA-oriented provi-
in the previous research work [9]. It highlights the sioning architecture for making it a Quality of
demand of negotiation framework establishment in Service-aware internet [17]. Later, end-to-end Qual-
the service layer of cloud (available on top of appli- ity of Service contract negotiation was introduced
cation layer). In futuristic cloud computing vision, the among network service providers using the dis-
negotiation framework is addressed as an imperative tributed reinforcement [18] and Q-learning algorithm
architectural component of SLA-oriented resource for their long-term benefits. The Cloud Comput-
management [10, 11]. This scenario motivates the ing Background Key Exchange [19] and HiTrust
research work towards the development of negoti- negotiation model is proposed to provide a secure
ation framework which can satisfy the customized scheduling and trust relationship among the dis-
cloud service provisioning mechanism. In addition, tributed service, respectively. In the communication
the cloud technology accomplishes the service dis- perspective, the negotiation framework can be clas-
covery, scaling, monitoring, and decommissioning sified into direct (straight forward) negotiation and
operations in dynamic manner, there is a demand indirect (intermediate third party) negotiation. The
for automated cloud service negotiation between con- direct type is preferred for single market (simple)
sumer and provider [12]. negotiation with two parties (service consumer and
R. Rajavel et al. / ANFBL strategy for effective decision making 2313

service provider), whereas an indirect model is used [40], Co-evolutionary learning [41]. Moreover these
for multiple inter-related market (complex) negoti- strategies can be applicable to the environment, where
ation with three parties (service consumer, broker, the opponent’s behaviors are predictable or precisely
and service provider) [20]. In direct negotiation, there known. Hence, there is lack of negotiation strategy
is a point-to-point negotiation process between the with learning ability, where the opponent’s behavior
service provider and service consumer. In case of is partially and imprecisely known (such as uncer-
indirect negotiation, a third-party is used for the nego- tain information about the preferences of opponent
tiation process via a single or more than one trusted agent).
third-party broker. To the best of our knowledge, only very few
research works are available in the context of fuzzy
2.2. Negotiation strategy based negotiation strategy. An agent-based fuzzy
constraint directed negotiation model is developed
The negotiation strategy describes the negotiation for solving the problem of supply chain scheduling
process among participants using a standard commu- and planning without using any third party agent [42,
nication protocol and the operational behavior of the 43]. The same fuzzy based negotiation mechanism is
offer or counter-offer generated by the participant’s extended for establishing the SLA between the con-
algorithm at the negotiating end. A process of offer sumer and provider in cloud computing environment
or counter offer generation, selection, and evalua- [44]. To enhance the negotiation outcome (satisfac-
tion algorithm constitutes the negotiation strategy of tion level) of consumer and buyer, an automated
the participants [21]. The participants may attempt negotiation process is utilized with fuzzy inference
to generate attractive offers in order to influence system for generating offer and counter-offer based
their respective counterpart by providing a dynamic on the requirements and preferences of negotiat-
concession in offers without receiving a concession ing parties [45]. To further improve the negotiation
in counter offers. Most research work follows the outcome, evolutionary approach is used in the nego-
trade-off and concession-making algorithm for the tiation strategy for learning the new relaxed criteria
negotiation strategy [22, 23]. A novel time con- fuzzy rules [46]. An artificial intelligence approach is
strained SLA negotiation strategy is proposed to cope embedded into negotiation process to form an adap-
up with the limited time duration of the negotiation tive fuzzy logic strategy for learning the behavior
session [24]. In the negotiation strategy, concession of the opponent agent [47]. These strategies lack of
is given using the time, opportunity, and competition third party broker involvement and also lack of oppo-
function. Here, the time function follows the con- nents fuzzy behavior learning mechanism during
servative, conciliatory, and linear concession-making negotiation process. So, this research work focuses
strategy [25]. To support negotiation with incom- on developing broker based adaptive neuro-fuzzy
plete information, a co-evolutionary learning method behavioral learning strategy for enforcing behav-
is used in the negotiation strategy with two types ioral learning concept under uncertain information.
of Estimation Distribution Algorithms (EDAs) such Moreover, this research work includes the non func-
as conventional EDAs and novel Improved Dynamic tional parameters like service name attribute, service
Diversity Controlling EDAs [26]. requirement attribute, service negotiation parameter
The automated negotiation strategy can be devel- attribute, and functional parameters like total nego-
oped without learning ability or with learning ability tiation time, communication overhead, utility value,
depending on the environment [27]. Much research and success rate.
works are available in the former case of nego-
tiation strategy without learning ability where the 2.3. Negotiation protocol
negotiation attributes like price, time-slot, and speed
are constantly changes over the time [28–31]. In The traditional model of negotiation reaches the
the later case of negotiation strategy with learn- agreement through the exchange of offer and counter
ing ability, very few research works are presented offer between the parties using Contract Net Protocol
using behavioral learning approaches like Bayesian and Alternate Offer Protocol. In the Contract Net Pro-
learning [32], neural network learning [33, 34], tocol [48], one end of the negotiation party can send
bulk negotiation behavioral learning [35], adaptive the offer and the opponent party can either accept or
probabilistic behavior learning [36], Reinforcement reject the offer. The actual concept of negotiation is
learning [37–39], Evolutionary behavior learning not realized in this protocol, because the opponent
2314 R. Rajavel et al. / ANFBL strategy for effective decision making

is not sending any counter offer. So, to realize the


negotiation concept, an alternate offer protocol [49]
is proposed where the opponent can accept, reject,
and modify the offer for sending the counter offer.
Later, Service Negotiation and Acquisition Protocols
were developed in the grid environment for support-
ing the reliable management of remote SLAs [50].
In order to establish the complex multi-party and
composite agreement, a requirement-driven negoti-
ation protocol is developed by iteratively discovering
the dependencies of the opponent’s requirements and
offers [51]. An automated negotiation is developed
in dynamic environments under restricted negotia-
tion time and interdependencies with three protocols:
Combinatorial Action Protocol, Cluster Bidding Pro-
tocol, and Mediated Negotiation Protocol [52]. To
automate the cloud service composition, a Focused
Selection Contract Net Protocol is used for dynam- Fig. 1. Fuzzy-based cloud service negotiation framework.
ically selecting the list of cloud agents and their
services [53]. A machine learning concept is inte- takes place between the ITBA and SPAs and finally
grated in Wrap-5 cross-layer routing protocol [54]. suggest the committed service to SCA. During the
According to recent literature study [55] and past lit- automation of negotiation process, the sequence of
erature study [56], an Alternate Offer Protocol is more offer and counter offer generated by the ITBA and
popularly used in all the real time applications due to SPA will exploits certain negotiation strategy at each
its significant realization and modeling of negotiation negotiation states. In the proposed framework, the
process during the implementation. ITBA exploits the adaptive neuro-fuzzy behavioral
learning strategy for maximizing the utility value and
success rate among the negotiation parties. This strat-
3. Proposed fuzzy-based cloud service egy will learn the behavior of the opponent at each
negotiation framework negotiation state and generates the adaptive counter-
offer. Also, it increases the negotiation coordination
A conceptual architecture of the proposed fuzzy- among the negotiating parties without causing any
based cloud service negotiation framework is negotiation conflict with the SPAs. Coordination may
represented in Fig. 1. The major components of leads to commitment of SLA and conflict may leads to
this framework includes service consumer, intelli- negotiation break off among the negotiating parties.
gent third party broker, service provider, directory Thereby, behavior learning will maximize the util-
facilitator registry service, Jade gateway agent ser- ity value and the commitment of SLA will maximize
vice, universal description discovery and integration success rate.
registry service, and adaptive neuro-fuzzy behavior
learning strategy. To automate the negotiation pro- 3.1. Adaptive neuro-fuzzy behavioral learning
cess, agent based technology is embedded to mimic strategy
the behavior of all the above components. Here, the
Service Consumer Agent (SCA), Intelligent Third- The conceptual architectural of adaptive neuro-
Party Broker Agent (ITBA), and Service Provider fuzzy behavioral learning strategy consist of several
Agent (SPA) are involved in negotiating the service components like offer sensor agent, control initiator,
on behalf of service consumer, intelligent third party sensorial data administrator, fuzzy behavioral learn-
broker, and service provider respectively. The SCA ing system, negotiation controller, basic negotiation
generates the requirements and preferences of ser- behaviors, and counter-offer actuator agent as shown
vice consumer and forward the same to intelligent in Fig. 2. All ITBA will invoke this strategy after
third party broker. This component will identify the receiving the negotiation offer from the opponents
appropriate ITBA to negotiate the service with mul- for the sake of generating appropriate counter offer
tiple SPAs. Now, the actual negotiation process will to the SPAs. An offer sensor agent always senses the
R. Rajavel et al. / ANFBL strategy for effective decision making 2315

initiator, a negotiation controller component com-


putes the activation levels of all behavioral rules
which are to be incorporated during the negotiation
states. Finally, it choose the appropriate levels of
behavior activation in the basic negotiation behav-
iors as a collection of operating functions with the
intention of operating the agent behavior in parallel.

3.2. Formulation of fuzzy based negotiation


process using reinforcement learning
and artificial neural networks
Fig. 2. Architecture of adaptive neuro-fuzzy behavioral learning
model.
The actual bilateral negotiation process between
the ITBA and SPA can be formulated according
to the adaptive neuro-fuzzy behavioral learn-
offer and gives perceived negotiation state informa- ing strategy using reinforcement learning and
tion to the sensorial data administrator and control artificial neural network techniques. During the
initiator for further analysis. A control initiator com- negotiation process, sequence of offer ρa→b and
ponent supplies the precision value of several binary counter offer ρb→a are generated by the broker
predicates (or switching of agent’s mission) based agent a ∈ {ITBA1 , ITBA1 , . . . ITBAn } and provider
on the computations of sensorial observations from agent b ∈ {SPA1 , SPA1 , . . . , SPAm } with respect
offer sensor agent. Next, sensorial data administrator  multiple negotiation attribute or issue i ∈
to
start aggregating the negotiation offers received from Quality Quality
iPrice
1 , iTime−slot
2 , i3 , . . . in . Here, the
sensors and matches with approximate behavioral
broker agents adopt the proposed behavioral learn-
rules available in the component of basic negotiation
ing strategy for generating the sequence of offers.
behaviors. Later, this matching behavior rules with
Similarly, the provider agents also adopt any one of
different activation levels has to follow by the ITBA
the negotiation strategy like trade-off, linear, con-
during the process of counter offer generation. Addi-
ciliatory and conservative concession for generating
tionally, the sensorial data administrator calculates
the respective counteroffer. At any time stamp tk ∈
precision values of the predicates exploited by the {t1 , t2 , . . . , tn }, broker agent a’s negotiation offer
rule-matching algorithm. Based on mission switch-
decision on provider agent b’s past counter-offer at
ing and matching behavioral rules information, the
time tk−1 is defined in Equation (1).
fuzzy behavior learning system will implements the ⎧ tk−1 tk−2
reinforcement learning capability by continuously ⎪
⎪ accept, if U(ρb→a ) ≥ U(ρa→b )

updating the rule base consisting of fuzzy context tk tk−1 tk−2
ρa→b = reject, if U(ρb→a ) < U(ρa→b ) (1)
rules (meta-rules). It operates on the inhabitants of ⎪

⎩ offer, otherwise
fuzzy rules that characterize the antecedent (behav-
ioral context) and the consequent (basic behavior) of k−2 t
the rule. Where, the total utility of the offer U(ρa→b ) or
tk−1
At each negotiation state, this system helps in counter offer U(ρb→a ) is computed with respect to
improving the speed of negotiation convergence number of attributes involved in negotiation pro-
by learning best behavioral context for any course cess as shown in Equation (2). In order to maximize
of basic behavior employed in opponent’s agent. the utility, negotiator can give different preferences
Moreover, the reinforcement learning evaluates the over the negotiation attribute by assigning appropri-
performance among the agents at the end of each ate weight to the attributes.
negotiation state and distributes the reinforcement 
n
t t
values to the rules that control the agent during nego- k−2
U(ρa→b )= Wp × u(ρa→b
k−2
[ip ] (2)
tiation process. Each rule is estimating the strength p=1
of the antecedent to represent the appropriate con- t
k−2
text for the application of basic behavioral activation Let u(ρa→b [ip ]) denotes the utility value of the
tk−2
by the consequent. According to sensed negotiation attribute ip over offer ρa→b , Wp denotes the weight
and current mission switching initiated by the control assigned to the respective negotiation attribute ip ,
2316 R. Rajavel et al. / ANFBL strategy for effective decision making

weight of each attribute range 0 < Wp < 1 and Table 1


thereby the total weight of attribute is equal to Fuzzy rule generation with respect to concession and preference
degree
one (W1 + W2 + . . . + Wn = 1). A value of each
fk∈{1,2} (ωPD) fk∈{1,2} (ωCD )
attribute range is [0,1] and it can be defined in terms
VL L M H VH
of fuzzy values as [ipVL , ipL , ipM , ipH , ipVH ]. Here,
VL FRVL,VL FRVL,L FRVL,M FRVL,M FRVL,VH
the offer attribute value VL, L, M, H, and VH stands L FRM,VL FRL,L FRL,M FRL,H FRL,VH
for very low, low, medium, high, and very high range M FRM,VL FRM,L FRM,M FRM,H FRM,VH
of values respectively. Normally, the highest util- H FRH,VL FRH,L FRH,M FRH,H FRH,VH
tk VH FRVH,VL FRVH,L FRVH,M FRVH,H FRVH,VH
ity value of one participant offer u(ρa→b pVH ]) =
[iITBA
1 may give lowest utility value for the oppo-
tk
nent u(ρa→b pVH ]) = 0. Similarly, the lowest utility
[iSPA Each dn impacts more on to the concession rate
tk
value of same participant offer u(ρa→b pVL ]) =
[iITBA derived from fuzzy rule FRf (ωPD ),f (ωCD ) , which
0 may give highest utility value for the opponent determines the different pair of membership func-
tk
u(ρa→b pVL ]) = 1.
[iSPA tions as states in Table 1. By exploiting these
During the bilateral negotiation process, the ITBA membership functions, one negotiating party can
only initiates the negotiation process by generating evaluate the behavior of opponent and then deter-
the offer proposal and also receives the correspond- mine the appropriate concession decision during
ing counter offer proposal from the SPA. From the counter offer generation process to quickly reach an
received offer, broker agent extracts the values of agreement. In order to handle dynamic situations of
negotiation attributes for computing the value of negotiation process, 25 different fuzzy rules adopts
opponent’s preference and concession degree by con- various concession rate according to five levels of
sidering the negotiation zone of the participants. membership functions. An average concession rate
Negotiation zone iZONE
p∈(1,n) is defined as the range of ωR can be defined by means of center of area method
negotiation attributes value between broker agent and for the defuzzification purpose as specified in Equa-
provider agent initial values, and agreement zone is tion (6).
defined as the range of negotiation attributes value
between broker agent and provider agent reservation
nd n × (ωR )n
values. Here, the initial value and reservation value ωR = (6)
represents the minimum and maximum value agreed n dn
to pay by the negotiating parties. At time tk the broker agent can generate the counter
tk
To effectively reach the agreement among the offer attributes ρa→b [iITBA
p ] based on received offer
negotiating parties, a concession degree and prefer- tk−1 SPA
ρb→a [ip ] at time tk−1 by adding or removing the
ence degree member functions are defined as shown
attributes range (iR 1 + i2 + . . . + in ) with average
R R
in Equations (3 and 4). This member function is used
concession rate as defined in Equation (7).
to predict the opponent’s preference degree (ωPD )
and concession degree (ωCD ) at each negotiation (iITBA
p )tk = (iSPA
p )
tk−1
⊕ iR
p × ωR (7)
state. Predicted values will be helpful during the
subsequent negotiation states for making appropriate Let the operator ⊕ denotes + and – for ITBA and
fuzzy behavioral decision by the proposed strategy. SPA respectively. The proposed negotiation strategy
generates the counteroffer by exploiting triangu-
1 + i2 + . . . + i n
iCR CR CR
lar shape to symbolize the fuzzy based preference
ωPD = (3)
iZONE
1 + iZONE
2 + . . . + iZONE
n
and concession degree membership functions with
respect to five levels as stated in Figs. 3 and 4
respectively.
1 + i2 + . . . in
iPR PR PR
ωCD = (4) In addition, the proposed negotiation framework
1 + i2 + . . . + i n
iR R R
defines the fuzzy behavior as triplet <C, D, E>. Let
The joint membership (desirability) function C be the negotiation context of application behav-
dn∈{1,2,3,4} can be expressed between two member- ior that describes the applicability of the behavior
ship function fk∈{1,2} (ωPD ) and fk∈{1,2} (ωCD ) as with respect to situations. D denotes the desirability
specified in Equation (5). function computed at each negotiation state-action
pair <s,a> that describes the desirability of perform-
dn = min[fk∈{1,2} (ωPD ), fk∈{1,2} (ωCD )] (5) ing action a at the negotiation state s of agent to
R. Rajavel et al. / ANFBL strategy for effective decision making 2317

behavioral activation is realized by the fuzzy rule as


shown in Equation (11).
If (negotiation state s IS in context C)
(11)
THEN (apply behavior A)

To make more evidence, distinguish global and


environmental context by rewriting the above rule as
defined in Equation (12).



negotiation state s IS in
Fig. 3. Preference degree membership function. If
environmental context EC
THEN (apply behavior A) (12)


negotiation state s IS in
WHEN
global context GC

Here the applicability conditions in the rule level


are treated as mission level equivalent of global con-
text. Therefore, replace the above rule as shown in
Equation (13).


negotiation state s IS in
If
environmental context EC
Fig. 4. Concession degree membership function. (13)
THEN (apply behavior A)
When (mission is M)
realize the basic ability. Finally, E represents the ele-
ment or object of negotiation environment on which Let M denotes the mission which occurs when the
the basic ability is applied. Enabling a fine compo- state s appears in the global context GC.
sition of desirability function D can be implemented The mission M is formally defined as 4-touple <G,
with respect to different behavioral activation at the AC, E, CS>. Where G is the goal of mission that
same time, by means of fuzzy rules as given in Equa- allows the agent to accomplish the task, AC denotes
tion (8). the applicability conditions in which mission have
If (negotitation state IS Si ) to be carried out, E is the element reference to the
(8) developed mission, and CS denotes the coordina-
THEN(action IS Ai )
tion of strategy among agents simple behavior that
Where each rule can be related with a function recognizes the mission. The basic idea of proposed
in terms of T-norms as defined in Equation (9). The adaptive neuro-fuzzy learning strategy is to formu-
desirability function of the negotiation behavior is late the model by applying the reinforcement learning
defined in terms of T-conorms as shown in Equation capability in artificial neural network in order to adapt
(10). the negotiation parameters and features of fuzzy logic
system. To further have the transparency, neural net-
DSi →Ai (s, a) = μSi (s) ⊗ μAi (a) (9) work model make use of rule-based fuzzy reasoning
during its construction. A major advantage of this
DB = ⊕i=1,2,...,n DSi →Ai (s, a) (10) model is to tune the rules of fuzzy behavioral learn-
ing system using appropriate behavioral decision and
Let n be the number of fuzzy rules that implements learning algorithm exploited in neural networks as
the basic behavior. The negotiation context of appli- shown in Fig. 5. The proposed fuzzy behavioral learn-
cation can be defined in terms of logical combination ing system consist of three layer feed-forward model
of predicates such as global context and environ- with input layer (crisp input), hidden layer (rule) and
mental context. Enforcing the negotiation context in output layer (crisp output). An input layer deals with
2318 R. Rajavel et al. / ANFBL strategy for effective decision making

Fig. 5. Neuro-fuzzy behavioral decision.

negotiation offers received by the broker agents dur-


ing the negotiation process. Hidden layer deals with
set of fuzzy if-then rules stated in Table 1. Finally,
the output layer represents the action aiming at gen-
erating the negotiation counteroffer decision at the
broker agent.

4. Experimental evaluation

The JADE simulation tool [57] is used to create


the experimental setup with set of negotiation par-
ticipants like SCA, SPA, and ITBA. Here, the real
scenario of this experiment setup is to simulate the Fig. 6. Negotiation preferences of ITBA and SPA.
e-commerce cloud service negotiation happening in
the multi cloud environment. Usually the actual bilat-
eral negotiation process will takes place between the
ITBA and SPA. So, the preferences of these agents After the simulation, the normalized utility value and
are given in the front-end portal to start the actual success rate are observed by varying the number of
negotiation process as depicted in Fig. 6. A bilateral negotiation rounds with respect to different negotia-
negotiation process between the agents can be visu- tion strategies as represented in Table 3.
alized through the sniffer agent as shown in Fig. 7. In It is clear from the comparison table, that
order to evaluate the proposed adaptive neuro-fuzzy the performance of proposed adaptive neuro-fuzzy
behavioral learning strategy against the existing strat- behavioral learning strategy outperforms the exist-
egy, this research work considered the parameters ing conciliatory and conservative strategies in terms
such as utility value and success rate of the negotiat- of utility value and success rate with respect to 50
ing participants. For effective validation, benchmark negotiation rounds. After increasing the negotiation
dataset used in the previous research study (Rajavel, rounds to 100, 200 and 500, the performance of pro-
2016) [35, 36] is considered as shown in Table 2. posed strategy further increases to maximum extent
R. Rajavel et al. / ANFBL strategy for effective decision making 2319

Fig. 7. Visualization of bilateral negotiation process using sniffer agent.

Table 2 due to the enforcement of a fuzzy behavioral learning


Experimental setting of e-commerce cloud service negotiation system model in the strategy.
process
The major advantage of the proposed ANFBL
Input Attributes Settings
strategy will give maximum utility value and suc-
Broker Provider
Agent Agent cess rate during negotiation process without any
Initial Price [10,60] [200,250] break-off among the negotiating participants. Only
Reserved Price [200,250] [10,60] for minimum 50 negotiation round the success rate is
Initial Time-slot [10,60] [300,350] average, after increasing the negotiation round more
Reserved Time-slot [300,350] [10,60]
than 100 will always gives the maximum level of
Negotiation Deadline [50,200] [50,200]
Rounds Rounds success rate. This achievement is possible due to
Negotiation Agent [5,20] [5,20] fuzzy based behavioral learning approach exploited
in the negotiation strategy. Whereas in the exist-
ing strategy various approach like conciliatory and

Table 3
Performance of negotiation strategies
ITBA vs. SPA Strategies 50 Rounds 100 Rounds 200 Rounds 500 Rounds
Utility Success Utility Success Utility Success Utility Success
value rate value rate value rate value rate
Conciliatory vs. conciliatory 0.049 0 0.098 0 0.196 1 0.490 1
Conservative vs. conciliatory 0.049 0 0.426 1 0.426 1 0.426 1
ANFBL vs. conciliatory 0.919 1 1 1 1 1 1 1
Conciliatory vs. conservative 0.491 0 0.983 1 1 1 1 1
Conservative vs. conservative 0.491 1 0.557 1 0.557 1 0.557 1
ANFBL vs. conservative 0.919 1 1 1 1 1 1 1
2320 R. Rajavel et al. / ANFBL strategy for effective decision making

conservative approach is used, which created con- management, Future Generation Computer Systems 29
stant rate of concession to negotiation offer or counter (2013), 472–487.
[5] M. Macias and J. Guitart, SLA negotiation and enforcement
offer. But, in the proposed ANFBL strategy conces- policies for revenue maximization and client classification
sion for the offer or counter offer is generated accord- in cloud providers, Future Generation Computer Systems
ing to the behavior of the opponent’s. Thereby, the 41 (2014), 19–31.
proposed ANFBL strategy can provide more utility [6] R. Rajavel and M. Thangarathanam, Optimizing negotiation
conflict in the cloud service negotiation framework using
value and success rate during the negotiation pro- probabilistic decision making model, The Scientific World
cess without any bothering of opponent’s negotiation Journal 2015 (2015), 1–16.
behavior. [7] R. Rajavel and M. Thangarathanam, A negotiation frame-
work for the cloud management system using similarity
and gale shapely stable matching approach, KSII Trans-
actions on Internet and Information Systems 9 (2015),
5. Conclusion and feature work 2050–2077.
[8] B.P. Rimal and E. Choi, A service-oriented taxonomical
spectrum, cloudy challenges and opportunities of cloud
The proposed research work evaluates the per- computing, International Journal of Communication Sys-
formance of negotiation frameworks with respect tems 25 (2012), 796–819.
to different negotiation strategies. This reach work [9] R. Rajavel and M. Thangarathanam, SLAOCMS: A lay-
identifies that the proposed adaptive neuro-fuzzy ered architecture of SLA oriented cloud management
system for achieving agreement during resource failure,
behavioral learning strategy outperforms the exist- Advances in Intelligent Systems and Computing 236 (2014),
ing conciliatory and conservative strategies in terms 801–809.
of utility value and success rate at the minimum [10] R. Buyya, S.K. Garg and R.N. Calheiros, SLA-Oriented
level of negotiation rounds. After increasing to max- Resource Provisioning for Cloud Computing: Chal-
lenges, Architecture, and Solutions, IEEE International
imum level of negotiation rounds, the proposed Conference on Cloud and Service Computing, 2011,
strategy always provides the utility value and suc- pp. 1–10.
cess rate due to learning capability embedded in [11] A. Corradi and O.F. Rana, Editorial: The management of
the strategy. To further increase the performance cloud systems, Future Generation Computer Systems 32
(2014), 24–26.
of the negotiation framework, the degree of trust [12] A.V. Dastjerdi and R. Buyya, An autonomous time-
worthiness and risk membership functions can be dependent SLA negotiation strategy for cloud computing,
added in the fuzzy behavioral learning mechanism. The Computer Journal 58 (2015), 1–15.
[13] S. Venticinque, L. Tasquier and B.D. Martino, Agents
The validity of the proposed negotiation framework
based Cloud Computing Interface for Resource Provi-
with respect to different combination of negotia- sioning and Management, International Conference on
tion strategies are demonstrated through exhaustive Complex, Intelligent, and Software Intensive Systems, 2012,
simulation experiments. In future, this negotiation pp. 249–256.
[14] K.M. Sim, Towards complex negotiation for cloud
strategy can also be extended with cognition and economy, International Conference on Advances in
evidence based behavioral learning approaches for Grid and Pervasive Computing, LNCS 6104, 2010,
further maximization of utility value and success pp. 395–406.
rate by minimizing negotiation conflict among the [15] S. Adabi, A. Movaghar, A.M. Rahmani and H. Beigy, Mar-
ket based grid resource allocation using new negotiation
participants. model, Journal of Network and Computer Applications 36
(2013), 543–565.
[16] I. Brandic, D. Music and S. Dustdar, Service mediation
References and negotiation bootstrapping as first achievements towards
self-adaptable grid and cloud services, Grids and Service-
Oriented Architectures for Service Level Agreements, 2010,
[1] A.G. Garcia, I.B. Espert and V.H. Garcia, SLA-driven pp. 119–132.
dynamic cloud resource management, Future Generation [17] D. Chieng, A. Marshall and G. Parr, SLA brokering and
Computer Systems 31 (2014), 1–11. bandwidth reservation negotiation schemes for QoS-aware
[2] R. Buyya, C.S. Yeo, S. Venugopal, J. Broberg and I. Brandic, internet, IEEE Transactions on Network and Service Man-
Cloud computing and emerging IT platforms: Vision, hype, agement 2 (2005), 39–49.
and reality for delivering computing as the 5th utility, Future [18] T. Groleat and H. Pouyllau, Distributed learning algorithms
Generation Computer Systems 25 (2009), 599–616. for inter-NSP SLA negotiation management, IEEE Trans-
[3] V.C. Emeakaroha, I. Brandic, M. Maurer and S. Dust- actions on Network and Service Management 9 (2012),
dar, Cloud resource provisioning and SLA enforcement 433–445.
via LoM2HiS framework, Concurrency and Computation: [19] C. Liu, X. Zhang, C. Yang and J. Chena, CCBKE-Session
Practice and Experience 25 (2013), 1462–1481. key negotiation for fast and secure scheduling of scien-
[4] M. Maurer, I. Brandic and R. Sakellariou, Adap- tific applications in cloud computing, Future Generation
tive resource configuration for Cloud infrastructure Computer Systems 29 (2013), 1300–1308.
R. Rajavel et al. / ANFBL strategy for effective decision making 2321

[20] K.M. Sim, Complex and concurrent negotiations for multi- [36] R. Rajavel and M. Thangarathanam, Adaptive probabilistic
ple interrelated e-markets, IEEE Transaction on Cybernetics behavioural learning system for the effective behavioural
43 (2013), 230–245. decision in cloud trading negotiation market, Future Gen-
[21] I. Rahwan, S.D. Ramchurn, N.R. Jennings, P. Mcburney, eration Computer Systems 58 (2016), 29–41.
S. Parsons and L. Sonenberg, Argumentation-based nego- [37] H. Pouyllau and G. Carofiglio, Inter-carrier SLA negotiation
tiation, The Knowledge Engineering Review 18 (2003), using Q-learning, Telecommunication Systems 52 (2013),
343–375. 611–622.
[22] X. Zheng, P. Martin and K. Brohman, Cloud Service [38] A. Rapoport, T.E. Daniel and D.A. Seale, Reinforcement-
Negotiation: Concession vs. Tradeoff Approaches, 12th based adaptive learning in asymmetric two-person bargain-
IEEE/ACM International Symposium on Cluster, Cloud and ing with incomplete information, Experimental Economics
Grid Computing, 2012, pp. 515–522. 1 (1998), 221–253.
[23] J. Yan, R. Kowalczyk, J. Lin, M.B. Chhetri, S.K. Goh and [39] A. Monteserin and A. Amandi, A reinforcement learning
J. Zhang, Autonomous service level agreement negotiation approach to improve the argument selection effectiveness
for service composition provision, Future Generation Com- in argumentation-based negotiation, Expert Systems with
puter Systems 23 (2007), 748–759. Applications 40 (2013), 2182–2188.
[24] G.C. Silaghi, L.D. Serban and C.M. Litan, A time- [40] E. Gerding, D.V. Bragt and H.L. Poutre, Multi-issue nego-
constrained SLA negotiation strategy in competitive tiation processes by evolutionary simulation, validation and
computational grids, Future Generation Computer Systems social extensions, Computational Economics 22 (2003),
28 (2012), 1303–1315. 39–63.
[25] S. Adabi, A. Movaghar, A.M. Rahmani and H. Beigy, [41] J. Gwak, K.M. Sim and M. Jeon, Novel dynamic diversity
Negotiation strategies considering market, time and behav- controlling EAs for coevolving optimal negotiation strate-
ior functions for resource allocation in computational grid, gies, Information Sciences 273 (2014), 1–32.
Journal of Supercomputing 66 (2012), 1–40. [42] C. Hsu, B. Kao, V.L. Ho, L. Li and K.R. Lai, An agent-based
[26] J. Gwak and K.M. Sim, An augmented EDA with dynamic fuzzy constraint-directed negotiation model for solving sup-
diversity control and local neighborhood search for coevo- ply chain planning and scheduling problems, Applied Soft
lution of optimal negotiation strategies, Applied Intelligence Computting 48 (2016), 703–715.
38 (2013), 600–619. [43] K. Kolomvatsos, K. Panagidi, I. Neokosmidis, D.
[27] M. Pooyandeh and D.J. Marceau, Incorporating Bayesian Varoutas and S. Hadjiefthymiades, Automated concurrent
learning in agent-based simulation of stakeholders’ nego- negotiations: An artificial bee colony approach, Elec-
tiation, Computers, Environment and Urban Systems 48 tronic Commerce Research and Applications 19 (2016),
(2014), 73–85. 56–69.
[28] S. Son and K.M. Sim, A negotiation mechanism that [44] L. Li, C.S. Yeo, C. Hsu, L. Yu and K.R. Lai, Agent-based
facilitates the price-timeslot QoS negotiation for establish- fuzzy constraint-directed negotiation for service level agree-
ing SLAs of cloud service reservation, Communications ments in cloud computing, Cluster Computing 21 (2018),
in Computer and Information Science 136 (2011), 1349–1363.
432–446. [45] B. Shojaiemehr and M.K. Rafsanjani, A supplier offer mod-
[29] J. Gwak and K.M. Sim, A novel method for coevolving PS- ification approach based on fuzzy systems for automated
optimizing negotiation strategies using improved diversity negotiation in e-commerce, Information Systems Frontiers
controlling EDAs, Applied Intelligence 38 (2013), 384–417. 20 (2018), 143–160.
[30] S. Son and K.M. Sim, A price-and-time-slot-negotiation [46] K.M. Sim, Evolving fuzzy rules for relaxed-criteria negotia-
mechanism for cloud service reservations, IEEE Trans- tion, IEEE Transactions on Systems, Man, and Cybernetics,
action on Systems, Man, and Cybernetics – Part B: Part B (Cybernetics) 38 (2008), 1486–1500.
Cybernetics 42 (2012), 713–728. [47] C.W. Khuen, C.H. Yong and F. Haron, A framework for
[31] S. Son and K.M. Sim, Adaptive and similarity-based trade- multi-agent negotiation system using adaptive fuzzy logic
off algorithms in a price-timeslot-QoS negotiation system in resource allocation, International Journal of Information
to establish cloud SLAs, Information Systems Frontiers 17 Technology 11 (2005), 35–49.
(2015), 565–589. [48] R.G. Smith, The contract net protocol: High-level com-
[32] D. Zeng and K. Sycara, Bayesian learning in negotia- munication and control in a distributed problem solver,
tion, International Journal of Human-Computer Studies 48 IEEE Transactions on Computers C-29 (1980), 1104–
(1998), 125–141. 1113.
[33] D.C. Moosmayer, A.Y. Chong, M.J. Liu and B. Schuppar, [49] S. Venugopal, X. Chu and R. Buyya, A negotiation mech-
A neural network approach to predicting price negotiation anism for advance resource reservations using the alternate
outcomes in business-to-business contexts, Expert Systems offers protocol, International Workshop on Quality of Ser-
with Applications 40 (2013), 3028–3035. vice, Enschede (2008), 40–49.
[34] G. Wang, T.N. Wong and C. Yu, A computational model for [50] K. Czajkowski, I. Foster and C. Kesselman, SNAP: A
multi-agent Ecommerce negotiations with adaptive negotia- protocol for negotiating service level agreements and coor-
tion behaviors, Journal of Computational Science 4 (2013), dinating resource management in distributed systems, Job
135–143. Scheduling Strategies for Parallel Processing 2537 (2002),
[35] R. Rajavel and M. Thangarathanam, ADSLANF: A nego- 153–183.
tiation framework for the cloud management system using [51] A. Klenk, A. Beck-Greinwald, H. Angst and G. Carle,
Bulk Negotiation Behavioural Learning approach, Turkish Iterative multi-party agreement negotiation for establishing
Journal of Electrical Engineering & Computer Sciences 25 collaborations, Service Oriented Computing and Applica-
(2017), 1–50. tions 6 (2012), 321–335.
2322 R. Rajavel et al. / ANFBL strategy for effective decision making

[52] F. Lang and A. Fink, Negotiating in dynamic environments: [56] V. Sanchez-Anguix, V. Julian, V. Botti and A. García-
Time efficient automated negotiations by means of combi- Fornes, Tasks for agent-based negotiation teams: Analysis,
natorial auctions, Evolving Systems 3 (2012), 189–201. review, and challenges, Engineering Applications of Artifi-
[53] K.M. Sim, Agent-based cloud computing, IEEE Transac- cial Intelligence 26 (2013), 2480–2494.
tions on Service Computing 5 (2012), 564–577. [57] M. Wooldridge, An Introduction to Multiagent Systems.
[54] B. Russell, M.L. Littman and W. Trappe, Integrating (2nd ed.). John Wiley & Sons, United Kingdom, Chapter
machine learning in ad hoc routing: A wireless adaptive 2, 2009.
routing protocol, International Journal of Communication
Systems 24 (2011), 950–966.
[55] B. Shojaiemehr, A.M. Rahmani and N.N. Qader, Cloud
computing service negotiation: A systematic review, Com-
puter Standards & Interfaces 55 (2018), 196–206.

Das könnte Ihnen auch gefallen