A Monitoring based Multi-Agent Filtering Approach for Web Service

Selection

Raja Bellakhal

, Fatma Siala

and Khaled Ghédira

National School of Computer Science, University of Manouba, Manouba, Tunisia

Higher Institute of Management, University of Tunis, Tunis, Tunisia

Keywords: Web Services Filtering, Multi-Agent System, Monitoring, Quality of Service (QoS), Negotiation, SOAP

Messages.

Abstract: During Web services selection processes based on the negotiation approaches, systems initially search for

services that comply with the users’ functional requirements. Then, based on the retrieved set of

functionally similar services, the negotiators start negotiation in order to come up with an agreement about

the QoS parameter preferences. Here the problem occurs when the number of services retrieved during the

first step is huge. In such case, the performance of the Web service selection process based on the QoS

requirements can be degraded. Examining the specifications of all retrieved services is certainly a waste of

time. Instead, it is more reasonable to remove all services that are unavailable and under the users’

requirement expectations before the start of the negotiation. To deal with this issue, in this paper we propose

a multi-agent based filtering approach that adopts a Web service monitoring method allowing the filtering

and the selection of the best candidate Web services for the selection process. The results of the conducted

experimentations demonstrate that adopting an agent-based filtering process decreases the CPU time of the

overall Web service selection process.

1 INTRODUCTION

With the growth in the number of functionally

similar Web services, efficient systems become

primordial to assist the users during the selection of

the suitable services that comply with their specific

requirements including the QoS parameters. Since

Web services are operating in dynamic and changing

environments, their QoS parameters change quite

frequently. Actually, these parameters are dynamic

and out of the providers’ control as for example, the

response time would depend on the bandwidth, the

transmission delay, the packet size, etc.

A series of Web service selection systems that

support the search for Web services based on the

users’ preferences in terms of the QoS parameters

has been presented. The dominating system among

them considers that QoS parameters are static

(Bentahar et al., 2008; Karray et al., 2013). Once the

quality of services is defined in the description and

published in the UDDI (Universal Description

Discovery and Integration) registry, they remain

unchangeable. In order to solve the problem of static

parameters, approaches based on statistic (Zhou and

Chen, 2009) network based techniques (Chen et al.,

2010; Benaboud et al., 2016) and negotiation

(Napoli et al., 2013; Linlin et al., 2013) are

proposed. Network and statistic based approaches

have the inconvenience of over resources

consumption, and the occurrence of conflicts

between the clients’ and providers’ preferences. The

negotiation has the advantage of solving

discrepancies among the clients’ and the providers’

conflicting preferences observed during the

aforementioned approaches.

The negotiation is a two-step process. During the

first step, the system searches for services that match

the clients’ functional requirements, while during the

second step, the best service is selected from the set

of the functionally similar services. A huge number

of candidate Web services selected during the Web

services discovery step can affect the selection

system performance during the selection step.

Imagine that during the second step, all the

discovered candidate providers go through many

rounds negotiation processes with the client.

Certainly, this will take a long time. Moreover,

implementing a negotiation system that simulates

Bellakhal, R., Siala, F. and Ghédira, K.

A Monitoring based Multi-Agent Filtering Approach for Web Service Selection.

DOI: 10.5220/0006900301450152

In Proceedings of the 14th International Conference on Web Information Systems and Technologies (WEBIST 2018), pages 145-152

ISBN: 978-989-758-324-7

145

the human real negotiation interactions and that

considers all discovered services is not an easy task.

The idea is to filter out the services that are

unavailable, and do not fit the users’ preferences. In

order to deal with the aforementioned problems, we

extend the negotiation system by adopting a filtering

approach based on a monitoring method. Before the

start of the negotiation process, the system must

search for the best candidate services.

The filtering process is based on monitoring

steps and evaluation of QoS parameters. Monitoring

of the Web services involves collecting information

about the real Web services performances during

runtime. After each monitoring step, a set of services

are filtered out based on their QoS values, and only

adequate services are retained.

The reminder of this paper is organized as

follows. Section 2 reviews related work. Section 3

describes the negotiation framework and the filtering

process. Section 4 presents and reports the

experimental results. Finally, the paper is concluded

in section 5.

2 RELATED WORK

There is a wide range of research around the Web

service selection systems based on QoS parameters.

Some of it is introduced to enhance the matching

techniques, the number of returned results, and to

optimize the selection system complexity. However,

these approaches consider that the QoS parameters

are static. Actually, Web services operate in a

changing environment, and systems designed to

discover them must consider the QoS parameters

dynamic. To tackle the problem of static parameters,

two series of approaches are proposed. The first

series is founded on network techniques and statistic

methods (Zheng et al., 2012), (Huang, 2013) and

(Benaboud et al., 2016). The disadvantages of these

approaches are mainly related to the over resource

consumption, the occurrence of conflicts between

the clients and the providers, and the lack of Web

service selection models. The second series is based

on negotiation models.

In (Napoli et al., 2013) a market based

negotiation mechanism among providers and users

that request a QoS aware service is presented. The

advantage of such a system is to consider the

dynamic aspect of the QoS parameters. The

providers may change the QoS values according to

their provision strategies. The authors describe the

negotiation as a bilateral process between two

agents. During the negotiation, agents exchange

offers and counter-offers. In the aforementioned

approach the negotiation could end without any

agreement since the process can terminate if a

deadline expires. Moreover, only providers are able

to generate offers whereas the clients can only

evaluate them. The contract net iterative protocol is

adopted. The authors note that this protocol has a

communication overhead, so its performance

degrades drastically. In (El-Awadi et al., 2014) the

authors present an SLA based negotiation approach.

The SLA contract must contain dynamically the

updated QoS values. In this work, a negotiation

engine is responsible for achieving an agreement.

Once the agreement is found, the monitoring system

checks whether the parameters related to the

retrieved services are equal, under, or over the

threshold defined in the SLA. Such a method may

not always guarantee an agreement between the

negotiators, since the generated SLA can be

withdrawn if the monitored QoS attribute does not

match the values defined in the SLA contract. In

(Bellakhal and Ghédira, 2016), an agent selection

system based on a hybrid negotiation is presented.

This system introduces new aspects that were

neglected by the existing systems. The first aspect

concerns the simulation of certain characteristics

observed during the real negotiation interactions

between negotiators. The second aspect relates to

making the negotiation dynamic by enabling the

negotiators to change their negotiation strategies

during the negotiation. The final aspect considers the

dependency between the concurrent negotiation

processes. The presented system is based on two

negotiation models, namely the argumentative and

the game-based negotiation models. In the presented

approach, a monitoring method that guarantees the

correspondence between the values generated by the

providers and the real Web services performance is

adopted. In (Ouadah et al., 2018), a hybrid approach

based on multi-criteria decision method to select the

best service from functionally similar services is

adopted. This approach is based on the reduction of

the decision space of the best candidate services for

the selection process. The adopted Web services

selection algorithm is based on aggregation of two

criteria namely, the user opinions about Web service

performance and the QoS parameters values

extracted from invocation history data.

The main issue with the aforementioned works is

related to the large number of candidate services that

are involved in the selection process. This may make

the process of finding the services that comply with

the specific users’ requirements longer and harder.

Moreover, in most of these works, the monitoring of

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

146

Web service performance is either conducted during

or after the selection process. On the one hand,

adopting a monitoring process during the selection

process can overload the system and postpone the

process termination. On the other hand, intercepting

the Web services performance after the selection

process termination can result in the process failure.

In this case, the real selected services performance is

different from the users’ preferences.

3 THE AGENT-BASED

FILTERING PROCESS

In this paper, we adopt an agent-based filtering

method over the set of discovered services in order

to come up with a final set of the best candidate Web

services for the agent-based negotiation process. In

the following, we start by describing briefly the

adopted negotiation framework as a basic element of

the proposed approach. Next, we focus on the main

objective of this paper, namely the presentation of a

novel filtering approach.

3.1 The Negotiation Framework

We adopt a multi-lateral negotiation framework

composed of concurrent sub-bilateral negotiation

processes. Several negotiation rounds are conducted

between competitive client agent’s instances and

provider agents. The client and provider agents are

in continuous competition in order to get the best

deal. They exchange offers and counter-offers while

adopting concession or trade off strategy. The

evaluation of the offers is based on the utility

function presented by Zheng et al., (Zheng et al.,

2012).

The adopted negotiation framework is based on

five agents namely, the provider agent, the client

agent, the intermediate agent’s instances, the client

agent’s instances and the coordinator agent.

 The provider agent

The provider agents involved in the

negotiation process represent the discovered

Web services. In the core of each provider

agent is implemented a negotiation model by

specifying its negotiation strategy and

protocol. The number of provider agents

depends on the discovered Web services. An

instance of the provider agent is run for each

retrieved Web services.

 The original client agent

The first role of the original client agent is to

provide to the client agent’s instances

information about the user’s preferences such

as QoS values and their related weights. Its

second role is to register offers resulting from

agreements between the provider and the

client agent’s instance.

 The intermediate agent’s instance

Dependencies among the concurrent sub-

bilateral negotiation processes are ensured by

the intermediate agent’s instances. They

communicate to the client agent’s instances

information about the offers of the competitor

providers. This information is exploited by the

provider in order to propose more interesting

offers for the client and get the deal.

 The client agent’s instance

The role of the client’s instances is to

represent concurrently the original client

agent’s behavior. They negotiate with the

provider agents by adopting different

negotiation models. Client agents are launched

as much as the provider agents. In the core of

each client instance agent is implemented a

negotiation model by specifying its

negotiation strategy and protocol.

 The coordinator agent

The first role of the coordinator agent is to

the client and provider agents. Its second role

is to inform the original client agent about the

offers resulting from agreements between the

negotiators.

3.2 The Filtering Process

Figure 1 depicts the filtering process and the agents

involved in it, namely the monitor agent, the

coordinator agent and the launcher agent. The

monitor agent has the role of intercepting

information about the real Web services

performance. This information is collected and used

by the coordinator agent in order to filter out

unavailable and inappropriate services. The

representation of the monitors as agents has the

advantage of gaining time. Once the monitor agents

are launched by the launcher agent, different

monitoring processes start by intercepting

simultaneously the Web services performance.

A Monitoring based Multi-Agent Filtering Approach for Web Service Selection

147

The filtering is a multi-step process based on a

multi-round monitoring method. The filtering

process and the monitoring processes are conducted

respectively in the core of the coordinator agent and

the monitor agents. The monitoring refers to the

control in runtime of the Web service performance

before the beginning of the negotiation process. This

has the effect of avoiding the overload of the system

by reducing the processing time. During each step, a

multi-round monitoring is adopted over the set of

services filtered during the previous step. Each Web

service belonging to the set of selected services will

be invoked according to λ rounds. The value of λ is

defined by the developer.

We focus on the selection of the candidate Web

services for the negotiation phase. We suppose that

SWS1 is the set of the initial Web services

discovered in response to the user’s request

expressed in terms of keywords and its initial QoS

requirements. The adoption of the keyword match

method during the discovery process seems too

simplistic here. Our main problem in this paper is

not to propose a full discovery and selection system,

but rather to study the impact of a filtering approach

over the overall CPU time related to the Web service

selection process.

The user’s requirements are defined by the

favorite QoS values and their related weights. The

set SWS1 will be the input of the first step in the

filtering process consisting in leaching out all

unavailable services. For this purpose, a monitoring

technique based on the SOAP

message is adopted.

Since QoS parameters are dynamic and change over

time, a monitoring process is essential for measuring

their values in real time. The QoS parameters are out

under the control of the provider. They depend on

the network bandwidth, the propagation delay, the

transmission delay, etc. Actually, we are not

responsible for hosting the Web services, so it is

impossible to control their performances. In order to

solve this problem, we adopt the SOAP message

calls. When a service is invoked, a request is sent to

the service in the form of a SOAP message and the

response is sent back to the client as a SOAP

message too. According to these request and

response messages, the values of two QoS

parameters, namely the response time and the

availability

are deduced. The equations (1) and (2)

are used to compute respectively their values.

http://www.soapuser.com/basics1.html

The Technical Guidance for the implementation of

INSPIRE View Services.

 Response time

The response time is the difference between

the time when the SOAP response is received

and the time when the user’s request is sent as

a SOAP message. Every time, a Web service

is called, the response time is computed.

Response Time = Time taken to complete the

response - Time taken for user request (1)

 Availability

Availability is the probability that the system

is ready for immediate consumption when

invoked. After a service call, when no

response is returned, the availability is set to

(2)

We also consider the price as a business parameter.

The price is proposed by the provider and we

assume that its value is constant.

n this paper, we focus on three parameters

namely, the response time, the availability and the

price. However, our system can be extended to

handle other parameters such as the reliability, the

integrity, and the accessibility of services. As is

mentioned in (Karthikeyan and SureshKumar, 2014)

these parameters can be estimated based on cost or

time. The monitor is a part of the Web service based

application. The system sends a service invoke, and

receives a reponse over the well-known XML and

SOAP messages. While invoking a given Web

service, the monitor records information about the

Web service invoking time.

When the service response is received, the monitor

records information about the Web service response

time, and the availability. A monitoring process is

launched over the services belonging to the initial

set SWS1. During the invocation of the Web

services, information about each service is recorded

in data storage. This information is used to deduce

the availability of each Web service. Based on the

computed availability values, the services are

classified into available and unavailable services.

The first step ends when all available services are

classified into the set SWS2.

unit

time

downtime

Availability =−

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

148

Figure 1: The agent based filtering process.

In the second step, the input of the filtering process

is the set SWS2, and its output is SWS3, the set of

services having a high and medium quality. In the

course of this step, services with poor quality are

filtered out. By definition, poor quality is related to

QoS parameters that are far from the users’

expectations expressed in terms of QoS parameters.

Since these parameters change over time and depend

on external factors, it is necessary to intercept their

values continuously. Thus, a second monitoring

process over the selected services belonging to the

set SWS2 is required. The problem here is set when a

single monitoring process round is adopted. In this

case, the choice of the services that must be removed

can be modified. A single monitoring process round

is when the service is intercepted only at a unique

time t. Actually, the QoS parameters change

frequently. During a small fraction of time, their

values continuously fluctuate. For example, we

consider two services WS1 and WS2. The response

time of each service is recorded at two different

times t1 and t2. The response time values of WS1

and WS2 at t1 are equal respectively to 2s and 6s.

However, at t2, the response time are 5s for WS1

and 3s for WS2. We conclude that at t1, WS1 is

better than WS2; however by comparing their values

at t2, WS2 performs better than WS1. So, when

applying the filetring process at t1, WS2 will be

filtered out although of it is better than WS1. To

avoid such mis-evaluation of the Web service

performance, a multi-round monitoring process is

needed to assess the real Web service performance.

Each service belonging to the set SWS2 is called

repeatedly in order to record and store the values of

the QoS parameters. These values are then

compared, sorted and used to deduce the minimum

and maximum QoS values related to each Web

service.

In order to compute the users’ utility generated

by the overall QoS parameters, we adopt the utility

function presented by Zheng et al., (Zheng et al.,

2012). A weighted sum function noted as U(m)

represented by equation 3 is adopted to compute the

general utility of a given offer m containing n

negotiation objects (QoS parameters).

() ()

Um W u x

=×



(3)

We consider that x is the value of a given QoS

parameter, and x

best

and x

worst

are respectively their

best and worst values. u

(x) the normalized value of x

is defined by equation 4.

()

worst

best worst

−

(4)

By comparing the utility generated by the maximum

and minimum QoS values with the user’s utility

generated by its preferred values, the system will

categorize the services according to their

performance into mediocre, medium and best

services.

We note by Min

resp

and Max

resp

respectively, the

minimum and the maximum values of the response

time recorded during the repetitive monitoring

process, while Min

avai

and Max

avai

are the minimum

and the maximum values related to the availability.

We also consider Val

resp

and Val

avai

the preferred

users’ values of respectively the response time and

the availability. We note by U

min

, U

max

and U

pref

the

user’s

utilities generated

repectively by the user’s

worst, best and preferred QoS values. Here, we

assume that the user’s utility increase (resp.

decrease) when the web service reponse time and

price values decrease (resp. increase) and when the

avalability value increase (resp. decrease). The

minimum utility of the client (U

min

)

will be equal to

the sum of the weighted maximum response time

and price values with the weighted minimum

availability value. Howerver, the maximum utility of

the client (U

max

)

will be equal to the sum of the

minimum weighted response time and price values

with the maximum weighted availability value.

The set of the mediocre (worst) services are

determined by comparing the utilities generated by

A Monitoring based Multi-Agent Filtering Approach for Web Service Selection

149

the users’ QoS preferred values with the utility

generated by the maximum response time and price

with the minimum availabiltiy value recorded during

the monitoring process. Here, we assume that the

price does not change during the monitoring rounds.

A service is a part of the set of the worst services

only if U

min

is lower than U

pref

. If the user’s utility

generated by the user’s worst values of QoS

parameters recorded during the monitoring is lower

than the utiliy generated by the preferred user’s

parameters, then the chance that such a service does

not match the user’s expectation in the future will be

high.

During the third step, if the performance of

services belonging to the set SWS3 is quite similar,

another criterion must be considered in conjunction

with U

min

and U

max

in order to distinguish between

the services.

The second criterion is related to the variation of

the QoS parameters values. The reliability of the

Web service depends on the variation of its QoS

parameters. If the variation of these parameters is

high, then the service is considered as unstable and

not reliable. However, when this variation is

minimal the service is qualified as stable and

reliable. The lower the variability of the QoS

parameters, the better the service is considered.

During a many rounds of monitoring process,

different values of response time and availability are

recorded. Many variation values coresspond for each

Web service. The worst variation is determined for

each service by comparing the different variation

values. The worst value matches the maximum

variation.

 The variation

The variation is the difference between a QoS

value recorded during a current monitoring

round, and another QoS value recorded during a

previous monitroing round.

We note by VarRt

resp

and VarRt

avai

the maximum

variation rate of respectively the response time and

the availability. Med

resp

and Med

avai

are the medians

of the maximum variation values respectively of the

response time and the availability. The services with

the worst variation values that are greater than the

median are considered of lower performance, while

the services with the worst variation values that are

lower than the median are of better performance.

The categorization of Web services into medium

and high quality depends on two factors namely, the

variation, the U

min

and U

max.

A Web service is ranked of a high performance

quality only if conditions 1, 2 and 3 are checked:

Condition 1: U

pref

is lower than or equal to U

min.

The

user’s goal is to increase its own utility. The first

condition requires that the utility generated by the

preferred response time, availability and price is

lower than or equal to the utility generated by the

maximum service response time and price as well as

the minimum service availability recorded during

the monitoring.

Condition 2: U

pref

is lower than or equal to U

max.

Condition 3: VarRt

resp

and VarRt

avai

are lower than

or equal to respectively the Med

resp

and Med

avai..

A Web service is regarded as having a medium

quality if condition 4 or condition 5 is checked.

Condition 4:

 U

pref

is lower than or equal to U

min

 VarRt

avai

is greater than or equal to Med

avai

Condition 5:

 U

pref

is lower than or equal to U

max

 VarRt

resp

is greater than or equal to Med

resp

4 EXPERIMENTATION

To get an in-depth investigation of the proposed

approach, we have implemented a negotiation

system considering the filtering process based on the

Java programming language and the multi-agent

platform MADKIT. We have deployed our system

in a Toshiba satellite L775 version with Intel(R)

Core (TM) i5 2410M CPU 2.3 GHz CPU and 4GM

RAM to simulate the environment. By these

experiments, we want to highlight three aspects.

First, we prove the role of the proposed filtering

approach in decreasing the CPU time of the Web

service discovery process. Second, we demonstrate

that an increase of the monitoring rounds has no

negative effect on the CPU time of the selection

process. Third, we prove the scalability of the

implemented system. In order to implement the QoS

monitoring process in the core of each monitor

agent, we adopt the SOAP message calls. We use the

framework SAAJ

(SOAP with Attachments API for

Java) in order to implement the SOAP message

calls. We design a user interface that enables the

user to sets its preferences in terms of the values and

weights of the QoS parameters as is presented in the

previous section. According to these initial

preferences, the set of functionally similar Web

services is selected. Information about the URL of

each service is extracted from the UDDI registry and

they are communicated to the launcher agent. The

http://docs.oracle.com/javaee/5/tutorial/doc/bnbhg.html

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

150

first role of the launcher agent is to create for each

Web service a monitor agent. Each monitor agent

has as attribute the URL of a Web service. The

second role of the launcher agent is to create an

instance of the coordinator agent. The monitor

agents start the interception of the Web services by

invoking them simultaneously.

In order to invoke a given Web service, an

instance of the soapConncetion class is created.

Then, the method call is applied over the latter

instance. The line code that represents the call

method is as follows: SOAPMessage soapResponse

= soapConnection.call(createSOAPRequest(), url).

The method createSOAPRequest() creates a SOAP

request message envelop which contains the HTTP

address of each Web service. This method returns a

SOAP message. When the SOAP response is

received, the printSOAPResponse(soapResponse) is

used in order to de-serialize the content of the SOAP

envelop. When a monitor agent terminates the

monitoring process, it sends the collected

information to the coordinator agent. The latter has

the role of making the filtering process or to launch

another monitoring process as is explained in the

previous section.

In the first series of tests, we set the number of

Web services and the monitoring rounds

respectively to 30 and 5. Here, the choice of the

number of monitoring rounds is independent of Web

services number. In Figure 2, we have reported the

CPU time produced by varying the user’s QoS

preferences in terms of the values and the weights.

In these tests, we consider three QoS parameters

namely, the price, the response time and the

availability. The results show that when the filtering

method is adopted, the values of the CPU time range

between 2.1 and 3.21 minutes. However, in the

opposite case the values exceed 5 minutes. In fact,

during the filtering, the services with QoS

parameters that are far from the user’s preference are

filtered out. In case the filtering is omitted, these

services can make the negotiation process longer.

Figure 2: Comparison of the CPU time (λ=5 and 30

available and unavailable WSs).

Indeed, the gap between the real Web service

performance and the user’s preference causes

conflicts and makes the negotiation process longer.

Moreover, during the normal negotiation,

unavailable services can result in an endless

negotiation process which prevents negotiators from

achieving an agreement. During the filtering process

such services are removed.

In the second series of tests, we change the value

of λ and we keep the same number of Web services

specified in the first test. We consider λ1=20. The

results are depicted in Figure 3. By comparing

Figure 2 with Figure 3 we deduce that the CPU time

increases slightly when the number of the filtering

rounds increases. However, it remains in most cases

lower than the CPU time recorded when the filtering

is ommitted. In fact, when the monitoring rounds

become higher, the filtering process takes more time

to come up with the best candidate services. In

overall, the increase of the filtering rounds has no

great consequences on the CPU time.

In the third series of tests, we launched our

system by varying the initial number of Web

services. We consider in the initial set of Web

services 10, 20 and 30 Web services respectively in

the first, the second and the third tests. The results

are reported in Figure 4.

Figure 3: Comparison of the CPU time (λ1=20 and 30

available and unavailable WSs).

We note that in the case where 10 Web services are

considered, the CPU time values range between 0.5

and 1.89 minutes. When 20 Web services are

considered, the CPU time values range between 1

and 2 minutes whereas, in the case of 30 Web

services the values range between 2 and 3 minutes.

We conclude that when the number of services

passes from 10 to 20, the CPU time was not greatly

influenced. However, when the number passes from

20 to 30, we notice that the CPU time has generally

increased by around 1 minute. From these results,

we conclude that the increase in the CPU time and

the Web services number are not proportional. This

A Monitoring based Multi-Agent Filtering Approach for Web Service Selection

151

means that an increase in the Web service number

by a given value x will not always result in an

increase in the CPU time by a fixed value y. Other

factors influence the CPU time such as the

performance of the selected services, the distance

between the user’s preferences and the Web services

QoS parameters as well as the number of unavail-

able services. The fewer the unavailable services are,

the better the CPU time will be. In the favourable

cases, when all services are available in the set of

initial Web services, the first step of filtering process

will be omitted. This will speed up the negotiation

process and makes shorter the CPU time.

Figure 4: The System Scalability.

5 CONCLUSIONS

With the rapid growth of Web services providing

functionally similar services, advanced discovery

systems based on the QoS parameters must be

adopted. The challenge is to come up with an

efficient system while keeping its generated results

reliable. Actually, the existing discovery systems

generate a huge number of candidate services for the

selection process. Considering all these services will

be a waste of time. To deal with this problem we

have presented a Web service filtering method based

on a multi-round QoS monitoring method. The idea

is to filter out services that are unavailable and under

the users’ expectations in terms of QoS

requirements. Actually the QoS parameters are

dynamic and change frequently depending on

external factors. Adopting a multi-round monitoring

process ensures an overview of the services

performance. In order to distinguish between the

Web service performance qualities, we introduced in

the filtering process the variation criteria. The

services that are characterized by a high fluctuation

of their QoS parameters are considered unstable

whereas those that have low fluctuations are

considered stable. Future work is related to

enhancement of the monitoring concept. In this work

we consider the monitor is part of the WS based

application. This can make sometime problems

when resources are limited and the requests cannot

be sent to the Web service. The latter issue needs to

be addressed.

REFERENCES

Bentahar J., Maamar Z., Wan W., Benslimane D., Thiran P.,

and Subramanian S., (2008), Agent-based communities

of Web services: An argumentation driven approach,

Service Oriented Comput. Appl., Vol.2, no. 4, pp. 219–

238.

Benaboud R., Maamri R., Sahnoun Z., (2016), PrefWS3:

Web Services Selection System Based on Semantics

and User Preferences, Informatica, Vol 40, No 2.

Bellakhal R., and Ghédira k., (2016), A multi-agent-based

negotiation system for web service selection. I. J.

Knowledge and Web Intelligence 5(4): 253-286.

Chen H-P., Zhang C., and. Yang G., (2010), A Model for

Managing and Discovering Services Based on Dynamic

Quality of Services. Journal of Networks, Vol 5, No 8,

pp.888-895.

El-Awadi R., Esam M., Rizka M., Hegazy A., (2014), A

Framework for Selecting Cloud Service Providers

Based on Service Level Agreement Assurance,

Proceedings of the International Conference on Grid

Computing and Applications (GCA).

Huang. X (2013), UsageQoS: Estimating the QoS of Web

Services through User Communities, ACM

Transactions on the Web (TWEB), Vol8 n.1, pp.1-31.

Karray A., Teyeb R., Ben Jemaa M., (2013), A heuristic

approach for web-service discovery and selection,

International Journal of Computer Science &

Information Technology (IJCSIT), Vol 5, Issue 2.

Karthikeyan, J & M, SureshKumar. (2014), Monitoring

QoS parameters of composed web services. 2014

International Conference on Information Communica-

tion and Embedded Systems, ICICES 2014.

Linlin. W, Saurabh. K. G, Rajkumar. B, (2013), Automated

SLA Negotiation Framework for Cloud Computing,

Cluster, Cloud and Grid Computing (CCGrid), 13th

IEEE/ACM International Symposium.

Napoli C. Di, Pisa P., and Rossi S., (2013). A market-based

negotiation mechanism for QoS-aware service selection,

The 15th International Workshop on Agent-Mediated

Electronic Commerce.

Ouadah, A., Hadjali, A., Nader, F. et al. (2018), SEFAP: an

efficient approach for ranking skyline web services,

Journal of Ambient Intelligence and Humanized

Computing, pp 1–17.

Zhou C., and Chen H., (2009), An Objective and Automatic

Feedback Model for Qos Evaluation, ICIS '09

Proceedings of the 2nd ICIS.

Zheng, X., Martin, P. and Brohman, K. (2012), ‘Cloud

service negotiation: concession vs. tradeoff approaches’,

Proceedings of the 12th IEEE/ACM International

Symposium on Cluster, Cloud and Grid Computing

(CCGRID), IEEE Computer Society, pp.515–522.

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

152