Continuous Service Optimization as Cloud Brokerage Service

Stamatia Rizou

, Yiannis Verginadis

and Gregoris Mentzas

European Projects Department, Singular Logic S.A., Athens, Greece

Institute of Communications and Computer Systems, National Technical University of Athens, Athens, Greece,

Keywords: Cloud Brokerage, Service Optimization, Adaptation.

Abstract: As cloud service ecosystems evolve, cloud service brokerage systems have emerged as intermediaries

between cloud service consumers and cloud providers to preserve and enhance user expectations. In this

paper, we describe the functionalities of a cloud brokerage tool that assist cloud brokers to ensure an

optimized cloud service selection both during cloud on boarding and operation phases. To this end, we

present an optimisation lifecycle providing customized continuous optimisation of cloud service selection

and we explain the role of the cloud broker within this service optimization loop.

1 INTRODUCTION

As cloud adoption increases, cloud service providers

broaden the range of their offered cloud services in

an effort to fulfill user expectations and stay

competitive in the cloud market. As a result, users

are offered with a wide spectrum of competitive

services that are difficult to evaluate and select. This

difficulty gives ground to the emergence of new

business roles in the cloud market landscape, such as

cloud brokers that act as intermediaries between the

cloud provider and the cloud consumer, to facilitate

cloud service selection and preserve the benefits of

the user.

In that respect, the problem of optimally

selecting cloud services can be seen as part of a

broader cloud brokerage system that aims to bridge

the gap between user expectations and service

provider’s offerings. Although existing approaches

tackle this problem by using multi-objective

optimization strategies and considering user

preferences (Garg, Versteeg and Buyya 2011), (Han,

et al. 2009), they have several limitations. First,

most of existing solutions have primarily focused on

the interaction between the cloud consumer and the

cloud provider without analyzing the involvement of

the cloud broker in the optimization process.

Secondly, existing works focus on the automatic

adaptation of services in Infrastructure as a Service

(IaaS) layer (Pawluk, et al. 2012), (Lawrence, et al.

2010) and do not analyze the effect of dynamic

conditions such as change in user preferences, price

etc. on the service optimization process. Moreover, a

number of multiple-criteria decision making

(MCDM) methods proposed in the literature (Garg,

Versteeg and Buyya 2011), (Han, et al. 2009)

capture user preferences by applying quantitative

techniques that cannot always depict the vagueness

of the qualitative aspects.

In this paper, we argue that the problem of cloud

service optimization needs to be addressed during

the whole service lifecycle across the different

service layers from IaaS to Software as a Service

(SaaS) layer. To this end, we introduce a holistic

approach that involves actively the cloud broker in

the optimization process and leverages optimization

goals to the specific application needs. Our approach

considers service optimization both during cloud

service on boarding and operation phases including

design as well as runtime service optimization.

2 RELATED WORK

Up to now, cloud services optimization has been

primarily investigated from a cloud provider's

perspective as a multi-objective constrained

optimization problem (Moon, Chi and Hacigümüs

2010), (J. Z. Li 2011). Although these approaches

aim to satisfy user requirements, they typically

consider user satisfaction as a constraint rather than

as the actual optimization goal. To this end, services

optimization problems in cloud brokerage systems

have the fundamental difference (with respect to the

382

Rizou S., Verginadis Y. and Mentzas G..

Continuous Service Optimization as Cloud Brokerage Service.

DOI: 10.5220/0004403603820385

In Proceedings of the 3rd International Conference on Cloud Computing and Services Science (CLOSER-2013), pages 382-385

ISBN: 978-989-8565-52-5

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

optimization problems from a cloud providers'

perspective) that they consider the fulfillment of user

expectations as the sole optimization goal.

From the perspective of the cloud consumer or

potentially the cloud broker, Han et al. (Han, et al.

2009) proposed a service recommender framework

using network QoS and Virtual Machine (VM)

platform factors for assisting user's decisions when it

comes to the selection of cloud provider. However,

in their work they do not consider user preferences

and they limit their evaluation criteria only to IaaS

specific factors. Closer to our goal, Pawluk et al.

(Pawluk, et al. 2012) have recently presented the

STRATOS cloud brokerage framework which

addresses the problem of dynamically selecting

resources from multiple cloud providers at runtime.

Furthermore, in (Lawrence, et al. 2010) the authors

propose the use of a so-called service optimizer (SO)

that continuously tests the compliance of service

execution with Service Level Agreements (SLAs)

through the use of dynamic SLAs. However, both of

these works focus mainly on the automatic cloud

service adaptation in IaaS layer.

An approach for cloud service ranking is

provided by SMICloud (Garg, Versteeg and Buyya

2011) which is a framework for comparing and

ranking cloud services. SMICloud is based on a set

of quantifiable measures, formally defined as SMI

attributes that model several quality dimensions of

cloud services. SMICloud uses an Analytical

Hierarchical Process (AHP) ranking mechanism to

solve the multi-criteria decision making problem of

finding the optimal cloud service. Similarly, Godse

et al. (Godse and Mulik 2009) applied an AHP

algorithm for ranking SaaS products.

Based on the above analysis, it is evident that

existing work has mainly focused up to now on the

optimization methodologies rather than on the

optimization process as a whole. Existing work on

the adaptation of service optimization (Pawluk, et al.

2012), (Lawrence, et al. 2010) mainly focuses on

automated service adaptation in IaaS layer and does

not consider the variety of changing conditions that

may occur in a cloud service ecosystem. Therefore,

our approach presented in the next section aims to

give an insight on the service optimization process

by proposing a continuous cloud service

optimization cycle, where cloud broker plays an

active role. Moreover existing optimization

methodologies consider only quantitative metrics by

assigning definite quantitative measures in user

preferences and in service characteristic evaluation.

Real world examples show that quantitative models

cannot always reflect the ranking among the services

accurately (Doyle and Thomason 1999). To this end,

in this paper we also motivate the need for

qualitative metrics to better model the imprecise

ranking among services.

3 SERVICE OPTIMIZATION

Our approach provides a holistic view on the

optimization process to help the user select the right

service. It includes the main design principles of an

autonomous system (Huebscher and McCann 2008),

i.e. it is based on a MAPE-K (Monitor, Analyse,

Plan, Execute, Knowledge) adaptation loop. The

lifecycle shown in Figure 1 consists of two different

concurrent iteration cycles, similar to the

visualization of adaptation mechanisms for service-

based applications in (Bucchiarone, et al. 2009). The

right cycle concerns the analysis, design,

development and deployment of the monitoring and

optimization mechanisms that constitute the control

layer which efficiently adapts the system in

continuous changes in the environment. The left

cycle applies the optimization and monitoring

mechanisms of the control layer by continuously

adapting the service provisioning layer. Both cycles

are applied both during cloud service on boarding,

i.e. initial cloud service migration as well as during

cloud service operation phase.

In more detail, the design time optimization

cycle sets the scope of the optimization with respect

to the application characteristics and the user

requirements. After the design of the appropriate

optimization and monitoring tools for the targeted

application, the optimization mechanisms will be

deployed in the system put into operational mode.

The optimization mechanisms are applied in the

continuous optimized service selection process

depicted in the left cycle. To this end, we have

identified four different steps that adapt the service

selection to current dynamic conditions. The same

four steps could be applied for the initial service

deployment, i.e. the cloud service on boarding as

well as for the continuous optimization of service

selection during the cloud service operation phase

after initial service deployment. Table 1 summarizes

the different information used before and after

deployment during the proposed four step process.

The proposed service optimization mechanism is

driven through the identification of optimization

opportunities. This step typically relies on the

automatic detection of the relevant information for

the optimization problem. In particular, during initial

cloud service on boarding, the optimization

ContinuousServiceOptimizationasCloudBrokerageService

383

Figure 1.

opportunities relate to the information that form the

conditions and input for the initial service

optimization, i.e. the early user requirements and

preferences and the historical data regarding the

service operation. Historical data are necessary for

the initial service optimization, since they constitute

the main driver for the evaluation of the candidate

services. In the case of the initial deployment, the

multi-objective optimization mechanism will assess

the candidate services through the use of Multi-

Criteria Decision Making (MCDM) methods, based

on the information presented in previous step. Note

that according also to related work (Garg, Versteeg

and Buyya 2011), (Godse and Mulik 2009), MCDM

is particularly appropriate for this kind of

optimization problem, since the assessment of the

candidate services is usually affected be several

different criteria that should be all considered before

making a decision. The multi-objective optimization

step will output a set of trade-off solutions that will

be subject to human decision making process.

During this step, the cloud broker plays an important

role in preserving user benefits during the

negotiation of service provisioning with the cloud

service provider. After the completion of the

negotiation step the initial service deployment is

realized. In that step, a cloud broker acts as an

auditor to guarantee the successful deployment of

the new service.

The cloud service optimization process during

the cloud service operation phase after the initial

deployment will follow a similar process. However,

the triggering events and adaptation strategies are

specific to the continuous service optimization

process that takes place after the initial service

deployment. In detail, the identification of

optimization opportunities at runtime refers to the

detection of events that trigger the service

optimization. These optimization triggers could be

either initiated by the user or by the cloud brokerage

monitoring mechanism which seeks for triggering

events tailored to the targeted applications. More

precisely user triggers could include changes in user

requirements which refer to changes in the required

quality of service or changes in user preferences

which refer to changes in the perception of “value-

for-money” and the prioritization of the several

service characteristics. In this respect, we assume

here that user experience after the service

deployment could lead to re-consideration of his

requirements, i.e. relax or stricken his QoS demands.

In addition to the user-initiated triggers, the

monitoring triggers of the envisioned cloud

brokerage mechanism could be the discovery of a

new service that could provide a better solution for

the user, changes in the price (e.g. offers, discounts

that could optimize the cost reduction for certain

users) or poor service performance that could refer

to SLA violations and performance degradation.

All the above triggers will lead to the calculation

of the new trade-off solutions with respect to the

different optimization criteria and subsequently in

CLOSER2013-3rdInternationalConferenceonCloudComputingandServicesScience

384

Table 1.

Life-Cycle Step Cloud service on

boarding phase

Cloud service

operation phase

Identify

optimization

opportunities

Early user

requirements, user

preferences,

Service operation

historical data

Monitoring events,

Changes on user

requirements,

preferences

Changes on service

provisioning (e.g.

offers, discounts)

Calculate

trade-off

solutions

MCDM based on

initial user

requirements and

preferences and

service historical

data

MCDM with up-to-

date user requirements,

preferences and

service monitoring

data

Negotiation

Price, SLAs,

commitment, service

characteristics

Price, SLAs,

commitment, service

characteristics , trust

relationship

Enact

adaptation

Initial service

selection and

deployment

Service replacement,

Service re-

configuration

the negotiation step. During the negotiation between

the cloud broker and the provider, prior user

experience with the service provider could lead to

different negotiation conditions with respect to the

case of the initial service deployment. Finally, the

optimization process at runtime will lead to the

adaptation of cloud services. The cloud broker is

again responsible for ensuring that the correct

actions, e.g. new service addition, service

replacement or service reconfiguration, takes place.

4 CONCLUSIONS

In this paper, we discussed the problem of optimal

cloud service selection in a cloud service ecosystem.

In particular, we presented a holistic approach that

describes all the steps of the optimization process in

a cloud brokerage system. Our proposal incorporates

the following concepts: (a) it involves the cloud

broker and the service user in the continuous

optimization loop, (b) it expresses user preferences

in an intuitive way that is understandable for the

user. In our short-term future work, we plan to

implement the proposed framework and validate its

correctness through extensive experimentation.

ACKNOWLEDGEMENTS

The research presented in this paper is supported by

the European Union within the FP7 Marie Curie

Initial Training Network “RELATE” and the FP7

ICT Broker@Cloud project.

REFERENCES

Bucchiarone, Antonio, Cinzia Cappiello, Elisabetta Di

Nitto, Raman Kazhamiakin, Valentina Mazza, and

Marco Pistore. “Design for Adaptation of Service-

Based Applications: Main Issues and Requirements.”

ICSOC/ServiceWave 2009. 2009.

Doyle, Jon, and Richmond H. Thomason. “Background to

Qualitative Decision Theory.” AI Magazine, 1999.

Garg, Saurabh Kumar, Steve Versteeg, and Rajkumar

Buyya. “SMICloud: A Framework for Comparing and

Ranking Cloud Services.” 2011 Fourth IEEE

International Conference on Utility and Cloud

Computing. 2011.

Godse, Manish, and Shrikant Mulik. “An Approach for

Selecting Software-as-a-Service (SaaS) Product.”

2009 IEEE International Conference on Cloud

Computing. 2009.

Han, Seung-Min, Mohammad Mehedi Hassan, Chang-

Woo Yoon, and Eui-Nam Huh. “Efficient service

recommendation system for cloud computing market.”

2nd International Conference on Interaction Sciences:

Information Technology, Culture and Human. 2009.

Huebscher, Markus C, and Julie A. McCann. “A survey of

autonomic computing\—degrees, models, and

applications.” ACM Computing Surveys, 2008:

40(3):1-28.

Lawrence, Andy, Karim Djemame, Oliver Wäldrich,

Wolfgang Ziegler, and Csilla Zsigri. “Using Service

Level Agreements for Optimising Cloud Infrastructure

Services.” ServiceWave Workshops. 2010. 38-49.

Li, Jim Zw and Woodside, Murray and Chinneck, John

and Litoiu, Marin. “CloudOpt: multi-goal optimization

of application deployments across a cloud.” 7th

International Conference on Network and Services

Management. 2011.

Moon, Hyun Jin, Yun Chi, and Hakan Hacigümüs. “SLA-

Aware Profit Optimization in Cloud Services via

Resource Scheduling.” SERVICES '10. 2010.

Pawluk, Przemyslaw, Bradley Simmons, Michael Smit,

Marin Litoiu, and Serge Mankovski. “Introducing

STRATOS: A Cloud Broker Service.” 5th IEEE

International Conference on Cloud Computing

(CLOUD). 2012. 891- 898.

ContinuousServiceOptimizationasCloudBrokerageService

385