AN ADAPTIVE REPLICATION FRAMEWORK

FOR IMPROVING THE QOS OF WEB SERVICES

Marwa F. Mohamed

, Hany F. ElYamany

Mohamed K. Hussien

Nashwa M. Yhiea

and Hamed M. Nassar

Computer Sciences Department, Faculty of Computers and Informatics, Suez Canal University, Ismailia, Egypt

Mathematics Department, Faculty of Science, Suez Canal University, Ismailia, Egypt

Keywords: Adaptive Replication, Quality of Service (QoS), Dynamic Load Balancing and Web Services.

Abstract: This paper presents an adaptive framework for managing dynamic replication of Web services in a

distributed environment including the Service-Oriented Architecture (SOA) environment. The framework

aims to improve the web services availability and to reduce the response time by supporting an automatic

replication of the consumed web services according to environment changing conditions that might occur at

the services provider side such as failure or increasing loading. For example, if one service or server fails,

the framework replicates automatically the consumed service on another particular selected server based on

some Service-Level Agreements (SLAs) including their performance and availability. Further, the

framework balances the incoming requests using Round Robin a load-balancing algorithm. Moreover, the

proposed framework is designed to predict the load of the involved candidate servers within the replication

process through utilizing a statistical regression technique.

1 INTRODUCTION

The Web service technology has gained a great

momentum in both academic and industry over the

last decades. A web service is a well-defined

software component which provides a specific

functionality over the Internet. Web services are

published, described, discovered and executed using

several standard protocols including WSDL for

service description, SOAP for intercommunication,

and UDDI for publishing and discovering the

abstracted and loose coupled web services

(Papazoglou, 2007). The quality of service (QoS) is

an essential issue for managing the access agreement

among the services providers and consumers when

selecting and composing the distributed service-

based applications. In particular, the Service Level

Agreement (SLA) is the predefined expected QoS

attributes standard including response time and

availability for controlling and monitoring the

interoperability between the services provider and

consumer. Specifically, response time is the time

required to complete a web service request.

Availability is the probability that the system will be

able to respond to the client request at any times

(May, Schmidt and Thomas, 2009).

Replication techniques can generally support web

services availability through providing multiple

replicas of web services; for example, if one service

fails, other copied services will be possibly initiated

in order to answer the client requests. Moreover, it

can basically reduce the response time through

broadcasting client requests among the available

replicas, particularly during runtime (Silva and

Mendonca, 2004).

In this paper, we will present a framework for

managing an adaptive replication of Web services in

a distributed environment such as the Service-

Oriented Architecture (SOA) environment. The

suggested adaptive replication framework supports

an automatic replication of different web services

according to the changing that might occur at the

services provider side or within the SLA elements of

the hosting servers. The proposed framework is

focusing on only replicating the web services, not

replicating the client requests. Basically, the

suggested framework issues the necessary decisions

when a service fails or overloads, and selects the

suitable servers that will be needed to replicate web

services on it automatically.

The rest of the paper is organized as follows.

Section 2 shows the related work. Section 3

283

F. Mohamed M., F. ElYamany H., K. Hussien M., M. Yhiea N. and M. Nassar H..

AN ADAPTIVE REPLICATION FRAMEWORK FOR IMPROVING THE QOS OF WEB SERVICES.

DOI: 10.5220/0003926102830286

In Proceedings of the 2nd International Conference on Cloud Computing and Services Science (CLOSER-2012), pages 283-286

ISBN: 978-989-8565-05-1

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

introduces a description of the suggested adaptive

replication framework. Section 4 describes the case

study and experimental results. Finally, conclusions

and future work are presented in section 5.

2 RELATED WORK

Several researchers have classified the process of

web service replication in different ways, for

example by determining which replica is handling

the incoming requests and selecting the optimal

server for hosting a replica. In (Salas, Perez-sorrosal,

Patiño-martínez and Jiménez-peris, 2006) the

authors classified the web service replication process

according to only determining which replica could

process the incoming requests. Basically, they

defined three different techniques called active,

semi-active and passive replication. In, (May et al.,

2009) the replication process has been classified

based on three techniques named as parallel, serial

and composite strategies. In (Zheng and Lyu, 2008),

a number of replication strategies have been

proposed based on adding time redundancy to a

combination of active and passive replication in

order to facilitate the process of selecting an

appropriate strategy for the different business case.

In (Ge and Zhang, 2010), a stochastic Perti net is

used to model the performance of composite

services as well as the environment in order to

achieve adaptive reselection during runtime. The

suggested framework accomplishes an adaptive

reselection of a higher priority service when the

failure occurs on the other services or an

environment. The service priority order is based on

availability of the network and the host machine;

and also the execution time and reliability of the

services. The main drawback of this framework is

that it does not consider some other crucial metrics

within the replica reselection process such as the

service or host load.

In (Yau, Goyal and Yao, 2005), an adaptive

replica selection framework has been proposed. The

framework considers both the characteristics of the

replica hosting environment and the consumer

requirements. The reselection process is based on

calculating a set of hybrid metrics including service

failure, host failure and dynamic load as well as the

customer business-value degree. However, the

framework has initially assumed that a fixed number

of hosts are available during runtime for the

replication process.

To this end, the main ideas of the proposed

adaptive service replication frameworks based on

selecting a web server for hosting the needed replica

from a predetermined set of available hosts (Ge and

Zhang, 2010; Keidl, Seltzsam and Kemper, 2003).

However, they didn’t take into account to predict the

changing load conditions of the service and the web

server during runtime.

3 THE ADAPTIVE

REPLICATION FRAMEWORK

This section introduces an adaptive framework for

adaptive replication management of Web services in

a SOA environment. The suggested adaptive

replication framework supports an automatic

replication of different web services according to the

changing that might occur at the services provider

side or the SLA elements of the hosting servers such

as capacity and availability. Generally, the

framework adaptively improves the availability of

the published services, and improves the Quality of

services (QoS) of the web services such as the

performance during runtime.

3.1 The Proposed Adaptive Replication

Framework

Figure 1 shows the suggested adaptive replication

framework. The framework involves three basic

layers: the clients, the replication middleware and

servers layer. The clients layer represents the

published services which consumers can access. The

servers layer contains the candidate hosts that might

be selected for holding possible replicas. Finally, the

replication middleware layer is located at the

services provider side.

Figure 1: The adaptive replication framework.

3.2 The Main Server Components

Sensors: This service is for monitoring all possible

CLOSER2012-2ndInternationalConferenceonCloudComputingandServicesScience

284

changes that can occur on the secondary servers,

such as: Failure sensor checks the secondary servers

status (available or failure). Performance sensor

collects information about the secondary servers as

the memory capacity, hard disk capacity and

processor type. Load sensor gets the CPU usage of

the secondary servers.

Selector: This service analyzes the information

received from the sensors. Then, it sorts the list of

secondary servers from the best to the worst

according to the following metrics as: server

Availability, server load and server capabilities.

Dispatcher: This service is the bridge between

the client and the secondary servers as it does the

following: Responder, it forwards requests from the

clients to the selected secondary server. Then, it

forwards response from the secondary server to the

clients. Load balancer, it executes the dynamic load

balancing process that demonstrated in (Alakeel,

2010).

Replicator: it replicates web services on the

selected secondary server. In case of the master

replica that process the client request fails or

becomes overloaded.

Cleaner: is acting as the garbage collector which

deletes unused web services from the secondary

servers.

4 CASE STUDY AND

EXPERIMENTAL RESULTS

One of the framework objectives is how to take

advantage of all available possible advantages in the

execution environment of any institution as much as

possible. For example, in a university environment;

the faculties differ in the number of student and the

date of grade announcement. We can see that one

faculty server is overloaded while others are less

loaded at the same time. Therefore, we thought

about how to take an advantage of the less loaded

servers when other servers become overloaded or

failure by proposing an adaptive replication

framework. In order to evaluate the proposed

framework, we present three different scenarios:

Static Scenario: The server of each faculty

works separately (or independently) and there is no

relation in between. For a student to check his mark

online; he can just type his ID and presses submit. If

the faculty server fails, the student will not be able to

check his mark. If the faculty server is overloaded,

the student will have to wait a bit longer to get his

mark.

Adaptive Replication Scenario: The university

server manages the web services replication process.

If the university server detects that one service or

faculty server fails, the framework automatically

replicates the consumed service on another

particular selected faculty server based on certain

Service-Level Agreements (SLAs) including their

performance and availability. Moreover, if the

university server detects that a particular service or

server is overload, the framework automatically

replicates the consumed service on another

particular selected faculty server then used the

Round Robin load balancing algorithm

(WebsiteGear, 2004) in order to balance the requests

between the replicas.

Adaptive with Load Prediction Scenario:

adding to the previous case, a prediction ability

using the following factors: The linear regression

model (Montgomery and Runger, 2007, chap 11),

and Scheduling which is the regular server schedule

that determines when it might be loaded.

The following experiments were conducted to

evaluate the performance of the framework using the

previous scenarios (static, adaptive and adaptive

with load prediction). The test environment is

VMware simulation consists of 6 servers: 1 MS, 4

SSs and 1 DNS. The capabilities of these servers are

established in Table 1. The experiments were

conducted using ApacheBench Version 2.0.40-dev

<$Revision: 1.146 $>apache-2.0

Table 1: The capabilities of simulation servers.

Capabilities MS SSs DNS

HD 40 GB 40 GB 40 GB

CPU Intel® core ™ Num of core using by

VMware 1

Memory 500 MB 384 MB 384 MB

OS Microsoft windows server 2003

Language PHP scripting language -

Figure 2: The average Throughput of running 7500

requests.

136912

Throughput(requestsper

second)

Concurrencylevels

static adaptive adaptivewithprediction

ANADAPTIVEREPLICATIONFRAMEWORKFORIMPROVINGTHEQOSOFWEBSERVICES

285

Figure 3: The average Response time of running 7500

requests.

In Figures 2 and 3, the performance order of the

previous approaches from the best to the worst is

adaptive with load prediction property, adaptive

replication and static approaches. Further, when the

concurrency level is increased, the gap between

static and adaptive is increased; and also the gap

between adaptive and adaptive with prediction

calculation is decreased except when concurrency

level is equal to 1.

5 CONCLUSIONS AND FUTURE

WORK

In this paper, a framework for an adaptive

replication process is introduced. The framework

includes several components for managing the

replication process in addition to a particular

component which is structured for predicting the

future load on the servers that host the original

copies of the web services. Further, a case study is

demonstrated as well as the experimental results are

provided. The experiment examined the suggested

framework within three different scenarios: static,

adaptive and adaptive with load prediction property

and measured the replication process performance

and throughput. The shown outcomes established

that the framework is working more efficiently when

it runs within the adaptive with prediction property

scenario.

For the future work, we plan to apply the

adaptive replication framework on composite web

services. Furthermore, we aim to apply partial

adaptive replication; the web services may

encapsulate more than one business operation and

the load on each operation is not equal. Hence, we

plan to apply an adaptive replication only on the

overload business operation. Finally, we also plan to

enhance the prediction load algorithm by using

another robust statistical prediction technique for

handling and broadcasting properly all data sets

types including the linear and nonlinear sets.

REFERENCES

Ali M. Alakeel, 2010. A Guide to Dynamic Load

Balancing in Distributed Computer Systems. In

International Journal of Computer Science and

Network Security.

Douglas C. Montgomery and George C. Runger, 2007.

Applied Statistics and Probability for Engineers, John

Wiley & Sons, Inc., USA, 4th edition.

Jorge Salas, Francisco Perez-sorrosal, Marta Patiño-

martínez and Ricardo Jiménez-peris, 2006. WS-

Replication: a Framework for Highly Available Web

Services. In 15th International Conference on the

World Wide Web. Edinburgh, Scotland.

Jose A. Silva and Nabor d. Mendonca, 2004. Dynamic

Invocation of Replicated Web Services. In the

Proceedings of the WebMedia & LA-Web 2004 Joint

Conference 10th Brazilian Symposium on Multimedia

and the Web 2nd Latin American Web Congress (LA-

Webmedia’04). Ribeirao, Preto, Brazil.

Liang Ge and Bin Zhang, 2010. A Modeling Approach on

Self-Adaptive Composite Services. In International

Conference on Multimedia Information Networking

and Security. Nanjing, Jiangsu China.

Markus Keidl, Stefan Seltzsam and Alfons Kemper, 2003.

Reliable Web Service Execution and Deployment in

Dynamic Environments. In 4thInternational Workshop

on Technologies for E-Services (TES). Berlin,

Germany.

Michael Papazoglou, 2007. Web Services: Principles and

Technology, Pearson Education Limited, England.

Nicholas R. May, Heinz W. Schmidt and Ian E. Thomas,

2009. Service Redundancy Strategies in Service-

Oriented Architectures. In Software Engineering and

Advanced Application. Patras, Greece. IEEE Comp.

Soc.

Stephen S. Yau, Gaurav Goyal and Yisheng Yao, 2005.

Replication for Adaptive Responsiveness in Service-

Oriented Systems. In International Conference on

Quality Software (QSIC’05). IEEE Comp. IEEE

Zibin Zheng and Michael R. Lyu, 2008. A distributed

replication strategy evaluation and selection

framework for fault tolerant web services. In IEEE

International Conference on Web Services. Beijing,

China.

WebsiteGear, Server Load Balancing: Algorithms, 2004.

Retrieved December 10, 2011, from http://content.web

sitegear.com/article/load_balance_types.htm.

100

150

200

250

300

350

136912

Responsetimeinmicrosecond

ConcurrencyLevels

static adaptive adaptivewithprediction

CLOSER2012-2ndInternationalConferenceonCloudComputingandServicesScience

286