2011 International Joint Conference of IEEE TrustCom-11/IEEE ICESS-11/FCST-11
Establishing Trust in Hybrid Cloud Computing Environments Jemal Abawajy, Senior IEEE Member School of Information Technology Deakin University, Melbourne, Australia
[email protected]
to the logical federation of autonomous computing clouds for the purpose of exchanging resources (storage, compute, messaging etc) in a uniform/unified way as a hybrid cloud computing [23]. In this paper, we refer to Grid-based distributed computing as public cloud computing whereas distributed computing provided by Google, Amazon, Microsoft and others that allows workloads to be deployed and scaled-out quickly through the rapid provisioning of virtual machines or physical machines as private cloud computing [23]. For example, Amazon’s Elastic Compute Cloud (EC2) [16] allows users to deploy VMs on demand on Amazon’s infrastructure and pay only for the computing, storage and network resources they use. The unprecedented introduction of the hybrid cloud infrastructure and services have allowed users to provision compute resources, storage resources, as well as applications in a dynamic manner on a pay per use basis. Although cloud computing provides a number of advantages that include economies of scale, dynamic provisioning, increased flexibility and low capital expenditures, it also introduces a range of new risks [23]. In particular, establishing trust relationships between different clouds in hybrid cloud systems is a fundamental and challenging research topic. This is because they are loosely coupled and more dynamic, which brings more challenges for the secure management and trustworthy collaboration. Therefore, the dynamic trust establishment problem between principals without pre-existing trust relationship becomes a bottleneck for cross-domain collaboration. Thus it is important to maintain a trust management system that permits one to distinguish the trustable from the non-trustable participants in a hybrid cloud model. Although, it has been clearly shown that an assurance of a higher degree of trust relationship is required to attain efficient resource allocation and utilization [6], to the best of our knowledge, the problem of how to determine service trustworthiness in hybrid cloud computing environments has not been addressed in a satisfactory manner. In this paper, we present a fully distributed framework that enable interested parties to determine the trustworthiness of hybrid cloud computing entities. Based on the framework, we propose an approach for detecting and filtering out dishonest feedbacks by using a personalized similarity measure to compute the credibility of feedback through personalized experience and a variable
Abstract—Establishing trust for resource sharing and collaboration has become an important issue in distributed computing environment. In this paper, we investigate the problem of establishing trust in hybrid cloud computing environments. As the scope of federated cloud computing enlarges to ubiquitous and pervasive computing, there will be a need to assess and maintain the trustworthiness of the cloud computing entities. We present a fully distributed framework that enable trust-based cloud customer and cloud service provider interactions. The framework aids a service consumer in assigning an appropriate weight to the feedback of different raters regarding a prospective service provider. Based on the framework, we developed a mechanism for controlling falsified feedback ratings from iteratively exerting trust level contamination due to falsified feedback ratings. The experimental analysis shows that the proposed framework successfully dilutes the effects of falsified feedback ratings, thereby facilitating accurate and fair assessment of the service reputations. Keywords-Cloud computing, management,Security .
I.
Trust,
Reputation,
Trust
INTRODUCTION
Cloud computing is a new distributed computing paradigm for on-demand and dynamic provisioning of computing resources to potential end-users leveraging virtualization and Internet technologies [24]. It delivers over the Internet infrastructure, platform, and software as subscription-based services. The computing services offered by cloud computing model are commonly referred to as Infrastructure as a Service (Iaas), Platform as a Service (PaaS), and Software as a Service (SaaS) respectively. These services are made available in a pay-per-use model to the potential cloud service consumers. Generally, cloud computing is classified into public, private and hybrid models. Public clouds are hosted and run by third parties and often shared by multiple users. It is highly likely that application from different users can simultaneously share cloud resources such as servers, storage systems, and networks. In contrast, private clouds are for the private use of an organisation and it is hosted at the organisation premises. This model gives organisations a high level of control over the use of cloud resources. Hybrid clouds combine both public and private cloud models. By enabling a private cloud to be augmented with the resources of a public cloud, the hybrid model can help to provide ondemand, externally provisioned scale. In this paper, we refer 978-0-7695-4600-1/11 $26.00 © 2011 IEEE DOI 10.1109/TrustCom.2011.18
118
A reputation system should have efficient representation of reputation as well as efficient mechanism for updating reputation and integrating efficiently the ratings of others. The quality of a reputation system depends on the integrity of the feedback ratings it receives as input. Due to the possible existence of subverted services, a trust entity for clouds faces the problem of integrating dishonest ratings. Thus, the challenge is how to systematically incorporate feedbacks collected from other clouds into computing trustworthiness of a given service. A fundamental problem is that a subverted cloud provider can rate an entity more positively or more negatively than the real experience with the agent would dictate. Thus, a false recommendation can result in committing a transaction with untrustworthy peers or avoiding a transaction with trustworthy peers. Therefore, effective protection against unfair ratings is a basic requirement and is an integral part of a robust reputation system.
tolerance threshold. We present the results of our evaluation of the proposed feedback filtering algorithms to show that it is effective in controlling the intrusion of dishonest feedback. The rest of the paper is organized as follows. The background and related work are discussed in Section II. We also discuss the problem of trustworthy resource selection and provisioning in inter-cloud computing environments. In Section III, the architecture of the Intercloud computing and the proposed trust framework models are discussed. In Section IV, we discuss the representation of reputation and how the reputation is built. We also discuss how reputation is updated as well as how the ratings of others are considered and integrated. In Section V, the design of experiments to quantify these concepts and evaluate them in terms of malicious behavior and interactions is unique in computer science. The analysis of the proposed algorithm is given. The conclusions and future directions are explained in Section VI. II.
PROBLEM STATEMENT AND RELATED WORK
B. Related Work Recent efforts have demonstrated interest in resource sharing across multi-site networks, such as Grids, in a coordinated manner. PlanetLab architecture is evolving to be deployed by other organisations and enable federations of PlanetLab’s [19]. Similarly, the Grid'5000 comprises nine sites geographically distributed in France [12]. Architecture and mechanisms based on the idea of peering arrangements between Grids to enable resource sharing across Grids is described in [2]. The focus of these prior works has been on the problem of resource exchange between different Grids. Trust is the firm belief in the competence of an entity to act as expected within a specific context at a given time [13]. Reputation is a measure that is derived from direct or indirect knowledge of earlier interactions of peers and is used to access the level of trust a peer puts into another [3, 13]. As an entity can trust another entity in the network based on a good reputation, we can use reputation to build trust [7]. This means reputation can serve, in the sense of reliability, as a measure of trustworthiness. Reputation management has an important role in establishing cooperative relationships between users and service providers by lowering some of the risks [3]. Trust can be used to measure our confidence that a secure system behaves as expected. A reliable trust management system provides capability to convert the unpredictable, highly dynamic pervasive environment into a trusted business platform. Thus, as the scope of hybrid cloud computing enlarges to ubiquitous and pervasive computing, there will be a need to assess and maintain the trustworthiness of the cloud computing entities. The reputation scheme helps building trust between peers based on their past experiences and feedbacks from other peers. Research in the area of trust and reputation systems has put a lot of effort in developing various trust
Trust plays an important role in hybrid cloud environments. It lends itself to estimate the trustworthiness of cloud service providers where trustworthiness could mean reliability, security, capability and availability. In this section, we will formulate the problem statement and then review related work. A. Problem Statement Overview In hybrid cloud computing, users and computational agents and services often interact with each other without having sufficient assurances about the behavior of the resource they entrust their data and applications with [23]. There are ample benefits for interconnecting computing clouds in a uniform way while respecting their autonomy. For example, the federated clouds will enable users to solve large-scale computational and data intensive problems in science, engineering, and commerce. These benefits have inspired research in creating mechanisms and protocols for interlinking exiting Grids across multi-site in a coordinated manner [2, 12, 18]. A federated system composed of autonomous distributed systems can pose several risks to the user communities. In intercloud environment, information such as resource provider's competence, honesty, availability, quality of service and reputation will influence the selection of the cloud provider to transact with. However, there is often insufficient information for deciding which resources to use. As the scope of hybrid cloud computing enlarges to ubiquitous and pervasive computing, there will be a need to assess and maintain the reputation of the entities. It is necessary to create a reputation manager that could capture and efficiently store the behaviour of entities, while being able to update it with new information if possible.
119
weaken trust management systems. The proposed trust model prevents many of such attacks and improves the reliability and the welfare of the system.
models and associated trust update algorithms that support users or their agents with different behavioral profiles. Trust and reputation systems have been recognized as playing an important role in decision making in the Internet world [7]. The recent work on trust management for Grid computing shows that modeling trust is of great importance for the future developments of the Grid computing [6]. As a result, integration of trust management system in standard grid computing has lately received attention [3, 6, 13]. For example, a trust brokering system that operates in a peer-topeer manner is proposed in [13]. An extension of Grid information service with reputation management service and its underlying algorithm for computing and managing reputation in service-oriented grid computing is discussed in [6]. An approach that enhanced the InetrCloud broker to evaluate the trustworthiness of the cloud service providers considering the user's feedback values and the performance criteria has been discussed in [19]. A method to filter out dishonest feedbacks for Bayesian reputation systems is presented in [21]. Bayesian reputation systems use the beta distribution in predicting a peer's reputation using the number of trustworthy and untrustworthy transactions as the distribution parameters. Honesty checking is performed by identifying feedbacks whose expected probability is less than the probability density function (PDF) at a certain quantile, q, and those exceeding the PDF at (1 - q) quantile in the beta distribution of the aggregated recommendations. The recommenders providing those outliers are considered to be dishonest. Abawajy [23] introduced an honesty rating factor that represents opinion about how credible ୨ אis as a provider of second-hand information. It is very difficult to distinguish between inactive or raters with fewer successful transaction from malicious raters. In the proposed approach, we develop a way to identify the credibility of the feedbacks and filter out those feedbacks that lie outside the norm. An approach for selecting computational Grid resources on the basis of trust and reputation to execute the jobs is discussed in [25]. A model based on controlled anonymity and cluster filtering methods to detect and exclude any highly unfair ratings is proposed in [27]. A collaborative filtering scheme is used to calculate an unbiased personalized reputation score. Using this method, groups of buyers who give similar ratings are classified into upper and lower classes. The final reputation score is calculated using the lower classes only. An approach based on the Dempster-Shafer theory of evidence to detect and protect users against spurious ratings is proposed in [26]. Although exiting works are complementary to the work proposed in this paper, to the best of our knowledge, this is the first work that attempts to build trustworthiness in intercloud computing. The focus on trust and reputation management systems is mainly to enhance the security of the web services. Moreover, exiting reputation systems for the standard Grids also suffer from a number of attacks that
III.
TRUST MANAGEMENT FRAMEWORK
In this section, we present a brief discussion of the main components of the inter-cloud computing architecture and the proposed mechanism for determining trustworthiness of a given resource. Let ܥଵ ǡ ܥଶ ǡ ܥଷ ǡ ǥ ǡ ܥ୬ be the universe of autonomous clouds. By autonomous we mean that no cloud has direct control and power over the actions of another cloud. For the purposes of this paper, we define a hybrid cloud computing as a subset of the universe of clouds consisting of ܥൌ ሺܥଵ ǡ ܥଶ ǡ ǥ ǡ ܥ ሻ. Cloud users require resources to deploy services and run applications. Cloud providers provide resources and services to potential users for fee or following another economic model such as bartering. Resource providers have their cost structures and policies that govern how their resources are provisioned to a user.
Figure 1: The hybrid cloud computing architecture Figure 1 shows the basic components of a hybrid cloud computing architecture that allows cloud users to deploy applications and scientific workflows that require resources beyond the capacity of their clouds. The architecture is layered in that services are provisioned to cloud users based on cloud-level by the cloud resource manager or the intercloud broker (ICB) level by two resource management policies.
120
•
Personal experiences: represents the first hand information of ୧ אafter transacting with ୨ א. Personal experiences are represented as୧ǡ୨ ൌ ۃǡ ୧ǡ୨ ;ۄ where ୧ǡ୨ denotes the personal satisfaction of ୧ א regarding ୨ אbased on the quality of the kth transaction. Personal satisfaction measure is continuous, between [0, 1] and its value is reflects 0 (not satisfied) and 1 (completely satisfied). • Reputation rating ( ୧ǡ୨ ) represents the confidence୧ א formed about ୨ ’ אs behavior as a good service provider. Trustors may specify a confidence in their own offered information items. This confidence is represented by a real number in the interval of [0, 1]. Similar to the trust certainty, 0 stands for no confidence whereas 1 stands for the highest possible confidence in the offered information. This confidence influences the trust update such that a weak statement with a low confidence leads to only a slight trust update, whether positive or negative. We assume that, from time to time, Clouds publish their personal experiences (୧ǡ୨ ) in the form of rating to a subset of their peers that they have a peering agreement with. The reputation rating ( ୧ǡ୨ ) is maintained privately while personal experiences (୧ǡ୨ ) can be shared with the members of the peering group in the form of personal satisfaction feedbackሺ݂ሻ as discussed in the next section. Furthermore, service reputation rating and personal experiences are separately updated after a transaction.
A. Cloud resource manager Users submit their resource allocation requests to their local Cloud Resource Manager (CRM). The CRM is responsible for the resource provisioning and allocation at the individual cloud level while the Virtual Infrastructure Engine (VIE) is the resource manager for the local cluster and can start, pause, resume, and stop Virtual Machines (VMs) on the physical resources. Resources sharing between multiple clouds to meet cloud user requirements are enabled by peering arrangements established between the participating clouds. The peering agreement describes the information that is to be exchanged under the terms stated in contracts such as SLA. The peering policy also describes the desired level of access control as well as mechanisms to protect data both in storage and transmission. B. Inter Cloud Broker The inter cloud broker (ICB) is responsible for mediating the resource exchange between peering clouds. ICB also provides external cloud selection capabilities by selecting suitable clouds that are able to provide the required resources to users’ requests. It is also responsible for managing requests for resources and services from other cloud ICBs. In addition, ICB is also responsible for monitoring the execution of applications across multiple clouds. The ICB also interacts with other entities including accounting systems that provide information on shares consumed by peering clouds.
IV.
FEEDBACK MANAGEMENT APPROACH
In this paper, we specify trust in the form of a trust relationship between two peering entities. This trust is always related to a particular context. Each cloud maintains and computes its reputation locally. Two algorithms have been developed to determine a user's trust value based on a sequence of her interactions and past behaviors.
C. Trust manger Trust manager enables cloud customers to select the best resources in the interCloud settings. It is responsible for getting the user's feedback, verifying the user's feedback and updating the same value in the feedback repository. Specifically, the trust manager is entrusted with the task of collecting and maintaining reputation rating, honesty rating and personal experience rating (i.e., first-hand information) about every other cloud provider that it has peering arrangement with. The rating is based on direct observation and experience (i.e., first-hand information) and indirectly by sharing observations and experience measures with other entities (i.e., second-hand information). In the proposed approach, entities can only share first-hand information. For example, a cloud ୧ אwill only share with a Cloud ୩ אits first hand experience about Cloud ୨ א. This allows us to avoid dependence and loops of reputation ratings. In this paper, we specify trust in the form of a trust relationship between two peering entities. This trust is always related to a particular context. Each cloud maintains and computes its reputation locally. The reputation manager uses the following ratings for building the reputation of an entity:
A. Personal experiences Personal experiences represent the first hand information of ݑafter transacting with ݓ. Personal experiences are represented as ୳ǡ୵ ൌ ۃǡ ୳ǡ୵ ۄ
(1)
where ୳ǡ୵ denotes the personal satisfaction of ݑregarding ݓbased on the quality of service in the kth transaction. Personal satisfaction measure is continuous, between [0, 1] where 0 (not satisfied) and 1 (completely satisfied). After finishing a transaction with a service provider, B, A provides to B either a rating of -1 (unsatisfactory) or 1 (satisfactory) depending on the outcome. Personal experiences are shared with the members of the peering group in the form of personal satisfaction feedback ሺ࣠ሻ . Specifically, cloud u can furnish personal satisfaction about cloud w as a feedback ൫࣠௨ǡ௪ ൯ to interested
121
parties on demand.Given the satisfaction rate of service u regarding service w’s performance in kth transaction (i.e., ܵ௨ǡ௪ǡ ) and the fading factor (i.e., Է௨ǡ௪ ), the personal satisfaction feedback scores ൫࣠௪ǡ௨ ൯ is computed as follow. ࣠௪ǡ௨ ൌ ݉݅݊൫ͳǤͲǡ σୀଵ Է௨ǡ௪ ܵ כ௨ǡ௪ǡ ൯
To measure the personalised similarity measure tolerance threshold ( ߩ ), Cloud peer w computes the feedback similarity between w and u over the common set of peers they have interacted with in the past as follows: σԳ సభ ௌೠǡ
ߩൌ
(2)
Է௨ǡ௪ ൌ ߜ
ȁ߬ ഥ ߬
V.
ೠǡೢ ାୖ౮אԳǡ౭
PERFORMANCE ANALYSIS
In this section, we evaluate the effectiveness of each of these heuristics. We will study the impact, effectiveness and cost of the different trust parameters in computing trustworthiness of a service. We will also study the reliability of the trust management is one of the important metrics to analyze the strength of a given trust management system.
(3)
B. Feedback filtering approach Reputation rating ( ୧ǡ୨ ) represents the confidence୧ א formed about୨ ’ אs behavior as a good service provider. Each cloud, upon request or periodically, provides feedback about the subjective quality of a received service from a given cloud provider. The metrics used to rate the quality of a service is a real number in the interval of [0, 1]. A perfect service is rated with 1 while 0 indicates a completely unsatisfactory one. The basic principle of the proposed approach is to use the personalized similarity measure to compute the credibility of feedback through personalized experience and a variable tolerance threshold. Specifically, when a customer w wants to evaluate the trustworthiness of a Cloud service provider u, customer w needs to compute the personalized feedback similarity between w and every other customer and filter out feedbacks from others that fail to meet the (personalised similarity measure tolerance threshold) condition. Let Գ be the set of Cloud customers that provided feedback for service . To measure the feedback score credibility of Cloud customer u, Cloud customer w computes the feedback similarity (ܵ݅݉௫ǡ௪ ) between w and its personal experience score as follows: หೠǡೢ ିୖ౮אԳǡ౭ ห
(5)
ୖ୩ǡ୨ ܵ݅݉௫ǡ௪ ߩ ୵ ൌ ቊ ୵ (6) ୳ǡ୵ Ͳ At the end of the transaction, Cloud customer w will send feedback rating updates to all peers that had rated Cloud customer u earlier.
Here, ߜȁͲ ൏ ߜ ͳ is a small positive constant and the parameter ߬ is the current time and ߬ ഥ represents the time that a given transaction is performed.
ܵ݅݉௫ǡ௪ ൌ ඨ
ඈ
In practice it means that if an Cloud customer's overall reputation score is outside the 1% quantile and the 99% quantile of the ratings provided by a rater, then that rater's ratings will be discarded from the set of ratings that contribute to the Cloud customer's score. Based on the similarity and threshold results, the Cloud customer w updates the reputation score of Cloud customer u as follows:
where n is the total number of successful transactions performed by w for u. When no experience with a service is made in a long time, the aged trust relationship might no longer be valid. This usually means that the trust confidence level decreases over time. The fading factor (i.e., Է௨ǡ௪ ) controls graceful reduction of personal satisfaction ratings as they age which means the recent personal satisfaction ratings are given more importance in computing feedback scores ൫࣠௪ǡ௨ ൯ based on personal satisfaction ratings. The fading factor (i.e., Է௨ǡ௪ ) is defined as a function of the number of successful transactions that service customer and the service providers had engaged in prior to computing the feedback score and is given as follow: ത ഓషഓ
ȁԳȁ
A. Experimental setup To assess the quality of the feedback filtering algorithm presented in the previous section, a series of test scenarios was developed. All simulations were conducted over 10000 sessions. The first 200 sessions of each simulation have not been included in graphs, since the market usually takes some time to stabilise as the reputations change to reflect the Cloud service provider's behaviour. To simulate various real-world reliability scenarios, we generate individual Cloud customer, we used a bimodal distribution to represent a system that has a mix of highly-reliable workers and compromised or poorly-connected nodes. We use feedback trust as a metric to evaluate the reliability of the proposed approach to measure the credibility of reporting ratings. Reliability of the trust management is one of the important metrics to analyze the strength of a given trust management system. A reliable trust management system should help the users to defend themselves against malicious information, including trust values propagated by other users into the system. The system is reliable if this property is accomplished.
మ
(4)
122
B. Reputation rating Figure 3 shows the impact, effectiveness and cost of the different trust parameters in computing trustworthiness of a service in the proposed approach. 0.9
tolerance threshold (ߩ), the less false positives (i.e. the less honest raters are excluded) and the more false negatives (i.e. the more unfair raters are included). The larger the value of ߩ, the less false negatives (i.e. the less unfair raters are included) and the more false positives (i.e. the more fair raters are excluded).
fading factor
0.8 0.7
personal satisfaction
0.7
0.6
0.5
0.5
tolerance threshold
Similarity Score
0.6 0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0 0
1
2
3
4 5 time
6
7
8
9
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 3: Fading weight impact. feedback Figure 4: Feedback filtering.
One of the targets of the proposed algorithm is to allow graceful reduction of satisfaction ratings as they age. When no experience with a service provider is made in a long time, the old trust relationship might no longer be valid. This usually means that the trust confidence level decreases over time. We represent the magnitude of this fading effect with a fading factor as a real number 0. A factor of 0 means no fading effect. The higher the fading factor, the faster trust relationship drops back to a state specified by the trust algorithm. This state might be a state of no trust and no confidence. Figure 3 (left) shows how a satisfaction rating fades with time. The significance is that the recent satisfaction ratings overweigh the past ratings.
The proposed reputation management system is reliable as it helps clouds to defend themselves against malicious information, including rating values propagated by other clouds into the system. The proposed reputation-based system is reliable as it minimises the dissemination of dishonest ratings by the peering clouds at various levels. Note that the second-hand information is the rating resulted from the bilateral direct interactions between the trustor and the trustee. Nodes without prior direct interactions cannot publish this information. This is because when a cloud ୧ א deliberates to enter a transaction with ୨ א, the two clouds complete and digitally sign an SLA contract. This ensures that both the service provider and the service user cannot deny entering into contractual agreement.
C. Feedback filtering Besides reputation rating, we also proposed a punishment mechanism to prevent liars from iteratively exerting bad behaviors in the system. The objective of this experiment is to measure the effect of the target Cloud customer’s behavior change on the performance of existing honesty checking mechanisms. Figure 4 shows the effectiveness of the personalized similarity measure to compute the credibility of feedback through personalized experience and a variable tolerance threshold. Feedbacks equal or less than the tolerance threshold are excluded from the consideration on service reputation. That is, the feedback from those peers with higher feedback trust ratings will be weighed more than those with lower feedback ratings. The result shows that our approach can effectively process strategic feedbacks and mitigate unfairness. The smaller the value of the personalised similarity measure
VI.
CONCLUSION AND FUTURE DIRECTION
In this paper, we presented a distributed reputationbased trust management system for hybrid cloud computing system. Trust value storage is distributed at the levels of the clouds in the system, which enables each cloud to make independent local decision for selection about trustworthiness of a cloud. Based on the trust management framework, we developed a mechanism that can effectively address strategic feedbacks and mitigate unfairness. We have studied the performance of the proposed trust management system in a simulated environment and due to space limitations this information is not fully provided here. Trust management is obviously an attractive target for adversaries. Besides well-known straightforward attacks such as providing dishonest ratings, some sophisticated
123
grid experimental testbed”. In 6th IEEE/ACM International Workshop on Grid Computing, 2005. [12] F. Azzedin and M. Maheswaran, “A Trust Brokering System and Its Application to Resource Management in Public Resource Grids”, in Proceedings of IPDPS 2004. [13] L. Peterson and J. Wroclawski. Overview of the GENI architecture. GENI Design Document GDD- 06-11, GENI: Global Environment for Network Innovations, January 2007. [14] Grid Interoperability Now Community Group (GINCG). http://forge.ogf.org/sf/projects/gin, 2006. [15] Amazon Elastic Compute Cloud [URL]. http://aws.amazon.com/ec2, access on Oct. 2009. [16] IBM Blue Cloud project [URL]. http://www03.ibm.com/press/us/en/pressrelease/22613.wss/, access on Oct 2009. [17] Rajkumar Buyya, Chee Shin Yeo, Srikumar Venugopal, James Broberg, and Ivona Brandic, Cloud Computing and Emerging IT Platforms: Vision, Hype, and Reality for Delivering Computing as the 5th Utility, Future Generation Computer Systems, 25 (6), pp: 599-616, 2009. [18] L. Peterson and J. Wroclawski, “Overview of the GENI architecture,” GENI: Global Environment for Network Innovations, GENI Design Document GDD-06-11, January 2007. [19] Vijayakumar, V., WahidaBanu, R.S.D., and Abawajy, J. Novel Mechanism for Evaluating Feedback in the Grid Environment on Resource Allocation, The 2010 International Conference on Grid Computing and Applications(GCA 2010), pp:11-17, July 12 - 15, Las Vegas, Nevada, USA. [20] Khan, K.M.; Malluhi, Q.; Establishing Trust in Cloud Computing, IT Professional, Volume: 12 Issue:5, page(s): 20 - 27, Issue Date: Sept.-Oct. 2010. [21] C. Dellarocas. Immunizing Online Reputation Reporting Systems Against Unfair Ratings and Discriminatory Behavior. In ACM Conference on Electronic Commerce, pages 150.157, 2000. [22] F. Azzedin, M. Maheswaran, and A. Mitra. Trust brokering and its use for resource matchmaking in public-resource grids. Journal of Grid Computing, 4(3):247–263, 2006. [23] Jemal Abawajy, Determining Service Trustworthiness in Intercloud Computing Environments, Proceedings of the 10th International Symposium on Pervasive Systems, Algorithms, and Networks (ISPAN '09), pp.784~788, 2009. [24] Yildiz, Mehmet, Abawajy, Jemal, Ercan, Tuncay and Bernoth, Andrew, A layered security approach for cloud computing infrastructure, Proceedings of the 10th International Symposium on the Pervasive Systems, Algorithms and Networks, pp. 763-767, [25] V. Vijayakumar, R. S.D. Wahida Banu & J. H. Abawajy (2011): An efficient approach based on trust and reputation for secured selection of grid resources,
attacks can undermine the whole trust evaluation process. We are currently developing a full list of threats against the proposed trust management and analyzing the vulnerability of the system to these threats. One of the main challenges in reputation system is to identify false feedbacks. The quality of a reputation system depends on the integrity of the feedback ratings it receives as input. Therefore, effective protection against unfair ratings is a basic requirement and is an integral part of a robust reputation system. Acknowledgement: The help of Maliha Omar, without whom this work will be difficult, is also appreciated. REFERENCES [1] Pinyol, I., Paolucci, M., Sabater-Mir, J., Conte, R.: Beyond accuracy. Reputation for partner selection with lies and retaliation. In: MABS, Honolulu, 15 May 2007 [2] Marcos Dias de Assuncao: Provisioning Techniques and Policies to Enable Inter-Grid Resource Sharing, PhD Thesis, Melbourne University, 2009. [3] Jemal H. Abawajy, Andrzej M. Goscinski: A Reputation-Based Grid Information Service. International Conference on Computational Science (4) 2006: 1015-1022. [4] Anthony Davison. Statistical Models. Cambridge University Press, Cambridge Series in Statistical and Probabilistic Mathematics, ISBN: 0521773393, October 2003. [5] S. Buchegger and J. Le Boudec, “A robust reputation system for p2p and mobile ad-hoc networks,” Proceedings of the 2nd Workshop on Economics of Peer-to-Peer Systems, 2004. [6] V.Vijayakumar and R.S.D.Wahida Banu, International Journal of Computer Science and Network Security, VOL.8 No.11, pp, 107-115, 2008. [7] A. Jøsang, R. Ismail, and C. Boyd. A survey of trust and reputation systems for online service provision. Decision Support Systems, 43(2):618–644, March 2007. [8] K. Miura, “Overview of japanese science grid project naregi,” Progress in Informatics, pp. 67–75, 2006. “Open Science Grid,” http://www.opensciencegrid.org, 2005. [9] L. Peterson, S. Muir, T. Roscoe, and A. Klingaman, “Planetlab architecture: An overview,” PlanetLab Consortium, Princeton, USA, Tech. Rep. PDN-06-031, May 2006. [10] C. Catlett, P. Beckman, D. Skow, and I. Foster, “Creating and operating national-scale cyberinfrastructure services,” Cyberinfrastructure Technology Watch Quarterly, vol. 2, no. 2, pp. 2–10, May 2006. [11] F. Cappello, E. Caron, M. Dayde, F. Desprez, Y. Jegou, P. Primet, E. Jeannot, S. Lanteri, J. Leduc, N. Melab, G. Mornet, R. Namyst, B. Quetier, and O. Richard. “Grid’5000: a large scale and highly reconRgurable
124
International Journal of Parallel, Emergent and Distributed Systems, DOI:10.1080/17445760.2011.575048. [26] B. Yu, M. P. Singh, and K. Sycara, "Developing trust in large-scale peer-to-peer systems”, 2004, pp. 1-10. [27] C. Dellarocas, "Immunizing online reputation reporting systems against unfair ratings and discriminatory behavior”, 2000, pp. 150-157.
125
