X-FEE

An Extensible Framework for Providing Feedback IN the Internet of Services

Anja Strunk

Department of Computer Science, Institute of Systems Architecutre

Dresden University of Technology, 01069 Dresden, Germany

Keywords: Internet of Services, Service-oriented Architecture, Web Services, Feedback, Monitoring, User Feedback.

Abstract: In the Internet of Services, there is a big demand for feedback about services and their attributes. For

example, on service market places, feedback is used by service discovery to help a service user to find the

right service or at runtime, feedback is employed to detect and compensate errors. Thus, the research

community suggests a large amount of techniques to make feedback available. However, there is a lack of

adequate feedback frameworks to be used to implement these techniques. In this paper we suggest the

feedback framework X-Fee, which is highly extensible, flexible and interoperable to easily realize feedback

components and integrate them in arbitrary infrastructures in the Internet of Services.

1 INTRODUCTION

The vision of the future internet is to transform the

web of information to a web of services. The goal is

to develop a so-called Internet of Services (IoS)

where economically viable services are offered,

brokered and consumed via the internet. Service

marketplaces emerge as web platforms where

providers can publish and sell services and

customers can find suitable services and combine

them together according to their needs.

While trading and using services in the IoS, a lot

of feedback information arises. The most important

are monitoring and user feedback. (Kalepu et al,

2002).

Monitoring feedback results from quality

monitoring. In the IoS, each service guarantees a

specific QoS level, such as receiving the service’s

response within a certain time period. The assured

QoS properties, negotiated between the service

provider and the service user are called Service

Level Objective (SLO) and saved in a legal contract,

called Service Level Agreement (SLA). At runtime,

quality monitoring observes the compliance of each

SLO by comparing the measured values with the one

that was promised. Thus monitoring feedback

deduces the service’s reputation related to their QoS

and their SLO compliance.

User feedback is created by human beings, after

each service interaction, by evaluating non-

measurable service attributes, such as correctness or

price-benefit-ratio. Hence it allows determining the

service’s reputation related to the non-technical

aspects.

The research community has suggested different

approaches to make both monitoring and user

feedback available in the IoS. In fact, there is a great

demand for feedback. For example, at design time,

feedback is consumed in order to improve existing

services or to deduce innovative ideas for new

services (

Stathel, et al, 2009). The service discovery,

a search engine for services, integrates feedback to

support the user by selecting the right service. (Xu et

al., 2007). At runtime feedback is used to recognize

errors and to trigger appropriate compensation

mechanisms (Strunk et al., 2009).

Despite the great demand for feedback and a lot

of techniques to provide it, there is a lack of an

adequate framework, which can be used to

implement these techniques. State-of-the-art

feedback components are not modular, interoperable

or flexible enough to be reused in new platforms.

This forces a costly re-development of already

existing software components.

In this paper we propose the feedback framework

X-Fee, which is highly extensible, flexible and

interoperable to be easily integrated in arbitrary

infrastructures. It acts as a base for the

implementation of methods for providing both

monitoring and user feedback.

119

Strunk A. (2010).

X-FEE - An Extensible Framework for Providing Feedback IN the Internet of Services.

In Proceedings of the 5th International Conference on Software and Data Technologies, pages 119-124

DOI: 10.5220/0002919401190124

 SciTePress

The rest of this paper is organized as follows.

Section 2 presents an overview of related work in

this area, followed by a description of the feedback

model used in X-Fee. The overall architecture of

X-Fee is explained in Section 4. In Section 5 we

discuss the strength and weaknesses of our approach

and finish the paper with a conclusion.

2 RELATED WORK

Feedback in service-oriented architectures and the

IoS is very well investigated. Started with the

question how feedback information can be

formalized, Maximilien and Singh (Maximilien and

Singh, 2002) published their conceptual model of

web service reputation. Today, this feedback model

is the de-facto standard and is also the basis for X-

Fee.

Kalepu et al. analyse feedback in the IoS and

categorize it into monitoring and user feedback

(Kalepu et al, 2002). With the author’s definition of

reputation as f(User Ranking, Compliance, Verity)”,

they argue that both feedback categories have to be

considered to get meaningful results.

The first approaches for techniques to provide

monitoring feedback are presented by Robinson

(Robinson, 2004) and Fickas et al. (Fickas et. el,

1995). Both authors propose a technique to monitor

web service requirements. Raimondi et al.

(Raimondi et al., 2008) investigate monitoring of

SLOs more closely and define a methodology for

online monitoring of web service SLOs based on

timed automata.

Beside techniques to monitor SLOs, commercial

and academic monitoring systems are developed.

However, the main disadvantages of these

approaches are the lack of interoperability,

flexibility and extensibility.

The well-know commercial monitoring system is

Nagios (Pervilä, 2007). Despite its many extensions

and modular sensors, it is not flexible enough to

monitor SLOs. Two meaningful academic monitor

approaches are Grand Slam (Spillner et al., 2009)

and SLAMon (Ameller et al., 2008). Both are

modular systems, which can be extended by future

functionality. Unfortunately the core modules can

not be adapted. Thus the fields of usage of Grand

Slam and SALMon are limited; for instance, the

used SLA description format can not be changed. On

the contrary, X-Fee allows modifying any part of the

system.

In contrast to SLO-monitoring, techniques and

systems related to user feedback are well analysed

and discussed. One of the best survey papers in this

area is the one by Wang and Vassileva (Wang and

Vassileva, 2007). The authors compare more than

ten academic as well as productive systems to

collect, compute and provide user feedback. None of

these approaches, however, is interoperable, flexible

and extensible enough to be used in heterogeneous

systems and to be adapted to individual requirements

of system providers.

3 FEEDBACK MODEL

Feedback in the context of the IoS evaluates a

service’s functional and non-functional attributes. It

is created for each service separately after each

interaction.

X-Fee organizes the different kind of feedback

information and the resulting reputations in a

conceptual feedback model, depicted Figure 1.

Figure 1: Conceptual feedback model.

The feedback model is based on the conceptual

model of reputation of Maximilien et al.

(Maximilien et al., 2002), which allows rating each

service’s attribute separately instead of the whole

service. This feedback information is defined as a 4-

tuple, consisting of the service and the service

attributes which were rated, the evaluation value as

well as the time stamp. All feedback information is

stored in the history, which is used to calculate

reputations based on reputation algorithms. A

reputation can be related to a service or just to a

service attribute. We call the first service reputation

and the latter attribute reputation.

The feedback model of X-Fee is independent of

the feedback category. Thus it can be used to save

monitoring as well as user feedback information.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

120

4 X-Fee – AN EXTENSIBLE

FEEDBACK FRAMEWORK

X-Fee is a modular framework based on OSGi

and

consists of five main parts: (1) a database to store all

feedback information, (2) the monitoring feedback

component, called SloMon, (3) the user feedback

component, called WebRat, (4) a web service

interface (WS) to support RPC-based access and (5)

a message-oriented-middleware (MoM) to support

event-based access to X-Fee.



Figure 2: The overall architecture of X-Fee.

As it was already mentioned, one of the main

disadvantage of state-of-the-art feedback

components is the lack of flexibility and

interoperability.

Interoperability is defined as the ability of a

software module to run on and communicate with

heterogeneous environments. X-Fee supports this

design principle as follows:

1. The OSGi container keeps X-Fee

independent from any target platform.

2. The web service interfaces of SloMon and

WebRat allow a programming language

independent RPC-based access to add and

get feedback as well as reputation values.

3. The message-oriented-middleware allows

SloMon and WebRat sending and receiving

XML-based events in a programming

language independent way.

http://www.osgi.org

Flexibility is defined as a system’s ability to be

adaptable to customers’ needs, by e.g. changing or

removing system’s modules. X-Fee supports this

feature as follows:

1. Both, the database as well as the message-

oriented-middleware are integrated via

adapters to make them replaceable.

2. Each part of X-Fee is implemented by two

separate OSGi bundles. One for the

interfaces and one for the implementation.

Hence, the customer can adapt the system

by exchanging the implementation bundle.

3. The modularity of OSGi allows using

SloMon and WebRat separately as well as

in an integrated way.

The vision of X-Fee is to provide a fully and

ready-to use feedback framework. Thus we are

realizing each part of X-Fee by a default

implementation. Therefore MySQL

is used as

database and the Apache ActiveMQ

as Message-

oriented Middleware (MoM). These open-source

third party components are not included in X-Fee

and have to be installed separately.

4.1 Monitoring Feedback with SloMon

SloMon, depicted in Figure 3, provides a framework

as well as a default implementation for SLO-

monitoring. Its name stands for SLO monitoring.

SloMon, which reuses the design pattern of Grand

SLAM (Spillner et al, 2009), consists of three main

parts: (1) the core, (2) the sensors and (3) the

aggregators.

The core provides the web service interface and

controls each part of SloMon. Therefore it is

composed of three services: (1) the agreement

service, (2) the sensor service and (3) the aggregator

service.

The agreement service is responsible for reading

agreements and for storing them into the database.

The default implementation is based on the

agreement standard Web Service Agreement

(

Andrieux et. al, 2007). Furthermore the agreement

service allows loading feedback values as well as

reputations from database via web service calls.

The sensor service manages the sensors, which

are responsible for providing the feedback

information. In terms of SLO monitoring feedback

information are related to QoS measurement and

SLO violation recognition, whereas each sensor

takes care for one QoS property. Both measured

http://www.mysql.com/

http://activemq.apache.org/

X-FEE - An Extensible Framework for Providing Feedback IN the Internet of Services

121

values as well as recognized SLO violations are

published on the MoM to be available for interested

components, such as error compensation strategies.



Figure 3: The architecture of SloMon.

SloMon supports two kinds of sensors: (1) periodical

sensors and (2) invoke sensors. A periodical sensor

measures the QoS property at regular intervals and

returns the value to the sensor service, which stores

the feedback information in the database. Therefore,

at each interval, the sensor service reads out all

agreements that are saved in the database and starts

for each SLO a new instance of the registered

periodical sensor, if such one exists.

An invoke sensor measures the QoS property

during service invocation. Therefore the sensor

service listens to the MoM to ServiceStarted events,

which have to be generated by the infrastructure

with which X-Fee is integrated. As soon as a

ServiceStarted event occurs, the sensor service reads

out all SLOs the started service guarantees and

activates the appropriate sensor instance. The

measurements are finished as soon as a

ServiceStopped event for the running service is

received on the MoM. The new feedback values

returned by the sensors are saved in the database.

By default SloMon supports two types of

sensors: (1) a response time sensor and (2) an

availability sensor.

The aggregator service is responsible for the

aggregators, which implement reputation

algorithms, such as a service’s availability per

specific time period, e.g. per day. Similar to the

sensors, each aggregator takes care of one reputation

value. SloMon distinguishes between two kinds of

aggregators: (1) service aggregators, providing

service reputations and (2) attribute aggregators

creating reputations for a specific attribute, i.e., QoS

property.

All Aggregators work in parallel and are

triggered by the aggregator service as soon as a new

feedback value was inserted in the database.

Therefore, the aggregator service reads out all

registered services as well as their related QoS

properties and starts a new instance of the

appropriate aggregator. The new instance gets the

list of all measurements related to the service or

service attribute, calculates the reputation value and

returns it to the aggregator service in order to be

stored in the database.

The default implementation of SloMon contains

two types of aggregators: (1) an availability

aggregator providing the service’s availability per

hour, day, week, month as well as year and (2) a

SLO violation prediction aggregator, which

forecasts the probability with which a service will

violates its SLO for response time and availability.

In contrast to the state-of-the-art monitoring

components, SloMon is built to be highly flexible

and extensible. Flexibility and extensibility allow

changing, removing and adding modules to adapt the

system to customer’s needs. SloMon supports these

features by implementing the services of the core

component as well as each sensor and aggregator as

separate bundles. This allows:

1. Changing the default implementation by

removing the old bundle and adding a new

one.

2. Adding a new sensors and aggregator, by

adding a new bundle implementing the

appropriate interface and registering the

sensor or aggregator at SloMon.

3. Removing unnecessary sensors and

aggregators by removing the respective

bundles.

Due to OSGi’s Hot Deployment, the adaptation

of SloMon is possible even if the system is running.

4.2 User Feedback with WebRat

WebRat, depicted in Figure 4, provides a framework

as well as a default implementation to deal with user

feedback in the IoS. Its name stands for web service

rating. WebRat consists of two main parts: (1) the

core and (2) the aggregators.

The core controls each part of WebRat and is

composed of two services: (1) the rating service and

(2) the aggregator service.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

122

The rating service implements a generic web

service interface to add new feedback information,

which means in terms of user feedback, to submit

user ratings for one or more service attributes. The

user ratings are stored in the database and distributed

in the MoM as XML messages in order to inform

components that are interested in. Furthermore it is

possible to access the reputations, related to services

and services attributes as well as to register services

and their rateable attributes.

Figure 4: Architecture of WebRat.

The aggregation service manages, similar to

SloMon, the service- and attribute-aggregators,

which calculate the reputations based on

appropriated algorithms. All aggregators work in

parallel and are started as soon as a new user rating

was inserted in the database. Hence, the aggregator

service reads out all stored services and their

attributes. For each service a new instance of each

registered service aggregator is started. For each

service attribute a new instance of the appropriate

attribute aggregator is activated, as long as such an

aggregator is registered at WebRat. Each aggregator

instance gets the list of user ratings stored for the

service or service attribute it manages. Based on this

list, the reputation is deduced and returned to the

aggregator service, which distributes the new value

in the MoM and inserts it into the database.

By default WebRat supports two kind of service

aggregators: (1) a simple arithmetic average and (2)

a reputation algorithm based on the decay effect (

et al., 2007).

Analogous to SloMon, WebRat is highly flexible

and extensible, by implementing the core

component’s services as well as each aggregator as

separate bundles. This allows:

1. Changing the default implementation by

removing the old bundle and adding a new

one.

2. Adding new aggregator bundle which

simply contains the algorithm to calculate

the reputation based on a list of user

ratings. All the other necessary tasks, such

as database or MoM access, are done by the

framework.

3. Removing unnecessary aggregators by

removing the respective bundles.

5 DISCUSSION

The vision of X-Fee, which is still under

development, is to be an interoperable, flexible and

extensible infrastructure for providing feedback in

the IoS. We fulfil theses requirements as follows:

1. Choosing OSGi as basis

2. Installing web service interfaces and MoM

3. Implementing all parts of X-Fee as

modules, which are as small as possible and

can be added, removed and changed

corresponding to the user’s needs

Despite the advantages discussed above, X-Fee

has also two main weaknesses: (1) the web service

interface and (2) the default implementation.

The web service interface constrains the

performance because of the overhead caused by the

XML marshalling and un-marshalling and the time

needed to establish and close HTTP connections for

every web service request. We are currently looking

for alternative ways to provide interoperable RPC-

based interfaces.

The second weakness, the default implement-

tation, provides only simple algorithms, which do

not conform to the current state-of-the-art and

missing important aspects, including security. Thus

the usefulness of X-Fee is limited without

addressing these issues. But the missing parts can be

implemented easily be replacing or adding new

bundles. We plan to provide X-Fee as an open-

source framework to be freely available.

6 CONCLUSIONS

X-FEE - An Extensible Framework for Providing Feedback IN the Internet of Services

123

This paper introduced X-Fee, an OSGi-based

framework to provide monitoring and user feedback

in the Internet of Services. X-Fee owns two main

components: (1) SloMon and (2) WebRat, which can

be used separately or in an integrated way to

implement techniques for feedback creation as well

as reputation calculation. In contrast to the state-of-

the-art feedback components, X-Fee is interoperable,

highly flexible and extensible and can be used by

heterogeneous applications. Moreover, it is adaptive

to the customers need.

ACKNOWLEDGEMENTS

The presented work is part of the research project

THESEUS

and will be evaluated in the use case

TEXO. The project was funded by means of the

German Federal Ministry of Economy and

Technology under the promotional reference

“01MQ07012”. The authors take the responsibility

for the contents.

REFERENCES

Ameller, D., Franch, X., 2008. Service Level Agreement

Monitor (SALMon). In Proceedings of the 7

IEEE

International Conference on Composition-Based

Software Systems (ICCBSS), Madrid, Spain, February

25-29, IEEE Computer Society, pp 224-227.

Andrieux, A., Czajkowski, K., Keahey, A., Ludwig, H.,

Nakata T., Pruyne, J., Rofrano, J., Tuecke, S., Xu, M.,

2007. Web Service Agreement Specification (WS-

Agreement). GFD-R-P.107, Open Grid Forum,

http://www.ogf.org/documents/GFD.107.pdf

Fickas, S., Feather, M. S., 1995. Requirement Monitoring

in Dynamic Environments. In Proceedings of the 2

International Symposium on Requirements

Engineering, York, England, March 27-29. IEEE

Computer Society, pp. 140.

Kalepu, S., Krishnaswamy, S, Loke, S. W., 2004.

Reputation = f(User Ranking, Compliance, Verity). In

Proceedings of the IEEE International Conference on

Web Services (ICWS’04), San Diego California, USA,

July 6-9. IEEE Computer Society, pp. 200- 207.

Maximilen, E. M., Singh, P. M., 2002. Conceptual Model

of Web Service Reputation. ACM SIGMOD Record,

SPECIAL ISSUE: Special section on semantic web

and data management. ACM, pp. 36-41.

Pervilä, M. A., 2007. Using Nagios to monitor faults in a

self-healing environment. Seminar and Self-Healing

Systems, University of Helsinki.

http://www.theseus-programm.de

Raimondi, F., Skene, J., Emmerich, W.:, 2008. Efficient

Online Monitoring of Web-Service SLAs, In

Proceedings of the 16th ACM SIGSOFT International

Symposium on Foundations of software engineering,

Atlanta, Georgia, November 9-14, CAM, pp. 170-180.

Robinson, W. N., 2003. Monitoring Web Service

Requirements, In Proceedings of the 11

IEEE

International Conference on Requirements

Engineering, Monterey Bay, California, USA,

September 8-12, IEEE Computer Society, pp. 65.

Spillner, J., Hoyer, J., 2009. SLA-Driven Service

Marketplace Monitoring with Grand Slam. In

Proceedings of the 4

International Conference on

Software and Data Technologies (ICSOFT09), Sofia,

Bulgaria, July 26-29.

Strunk A., Braun I., Reichert, S., Schill A., 2009.

Supporting Rebinding in BPEL processes. In

Proceedings of the IEEE International Conference of

Web Services (ICWS’09). Los Angeles, California,

USA, July 6-10, IEEE Computer Society, pp. 864-871.

Stathel, S., van Dinther, C., Schönefeld, A. 2009. Service

Innovation with Information Market. In Proceedings

of the 9

International Conference on Business

Informatics (Business Services: Concepts,

Technologies, Applications), February 25-27, Vienna,

Austria, pp. 825-834.

Wang, Y., Vassileva, J., 2007. Towards trust and

reputation based web service selection: A survey. In

Proceedings of the 27

International Conference in

Distributed Computing Systems Workshop (ICDCSW),

Toronto, Canada, June 22-29, pp. 25.

Xu, Z., Martin, P., Powley, W., Zulkernine, F., 2007.

Reputation-Enhanced QoS-based Web Service

Discovery. In Proceedings of the IEEE International

Conference on Web Services (ICWS’07), Salt Lake

City, Utah, USA, July 9-13, IEEE Computer Society,

pp. 249-256.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

124