Robust Reputation Based Trust Management ... - Semantic Scholar

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 21 (2016) pp. 10601-10605 © Research India Publications. http://www.ripublication.com

Robust Reputation Based Trust Management Framework for Federated-Cloud Environments M.N. Derahman, A. Abdullah and M. F. Azmi Faculty of Computer Science and Information Technology University Putra Malaysia. Abstract Recently, cloud computing is emerging drastically due to its ability to provide computing and other services to customers seamlessly. However, in heterogeneous environment, when there are various consumers and multiple service providers interacting and sharing resources, concern on security issues is crucial. Hence, the trust elements can be a motivating factor that can help to materialize the consumers on demand resource request transparently. It is worth to note that establishing trust is not an easy task especially when the attackers manipulate reputation feedback that is supposed to be trustworthy. Thus, the feedback mechanism based on the trust level needs to be further studied to increase reliability of trust management, particularly when involves with a largescale environment where strangers are competing with each other in offering services that are said to meet their quality of service and high reliable. In this study, we argue that the final trust can be manipulated by intruders using reputation feedback mechanism with untrustworthy value that inherited to the final trust value that can be wrongly interpreted by the consumers. To overcome this scenario, we suggest the rate of change to be used together with final trust value as it makes the trust value more robust where the intruder’s fake reputation feedback can be neglected. We prove this assumption using descriptive statistic where the standard deviation of proposed approach is low. As a result, the rate of change is significant in determining the trustworthy of the cloud service providers(CSP). Keywords: Computing.

Trust

Management,

Reputation,

Cloud

INTRODUCTION Cloud computing is an emerging computing paradigm enables resource sharing between participation entities. It is characterized by elasticity, flexibility and on-demand resource request for customers. The concept of utilities market namely pay-per-use seems to be nominated between the consumer and resource/service providers. Hence, most of the cloud providers have expanded their offerings to include compute-related capabilities such as storage, VM and also OS services. Collaborating anonymous entities will create challenges such as in security, privacy and trust management. Trust management can be defined as assurance and confidence level that people, data and objects will perform and/or behave in a projected manner. Trust is a conceptual relation between two entities of humans, machines and the communication between them [7]. In this perspective,

particularly in the cloud environment, we sometimes need to put our secret information and data to the third party facilities and infrastructure or allow others to access our facilities. Both are subjected to data manipulation and hacking even though the data is encrypted or our resources are protected. Thus, the trust element must play an important role. In federated-cloud environments, it involves with many entities across multiple domain of consumers, demand support from various service providers with specific properties and goals. Establishing trust between strangers is an important mean for cloud computing for the demand and supply chain to be materialised. Besides that, it needs to protect privacy and maintains the quality of service. Current studies among others are focusing on trust as a service, Service Level Agreement (SLA), feedback based trust management. This will be an issue where the particular participants are concerned on preserving consumer privacy, protecting services from attackers, maintaining service availability and allowing automatic service discovery. In this study, we propose an enhancement on trust management framework in federated cloud environment where it is solely upon to the trust results value. In this case, we focus on the resolution where there is the case that the attackers forging multiple identities and refute the final value of trust by falsifying its feedback reputation. It is formally known as Sybil attack [2]. Thus, the final trust value based on the accumulative results cannot be considered as only a determination factor in order to decide either the cloud service provider is trusted or not. This study focuses on how to nullifying those false reputations into the CSP. Hence, the Sybil attack will not significantly affect the final trust value. RELATED WORK In e-commerce application, maintaining the trust level between participants and share to the other potential customers is crucial. This is to develop the trust relation- ship in allowing resource sharing with many perspective, among them is pay-as-you-go using various platforms. Many studies have proposed a centralised node acts a broker with a trust management capability [6][3][1]. In [1], suggest the broker is connected with other six cloud nodes establishing a federated cloud infrastructure. Due to its one point of failure and subject to security ramification. The cloud infrastructure here solely depends on its intermediate node. Meanwhile, in [3] the framework is using trust mediator as a

10601

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 21 (2016) pp. 10601-10605 © Research India Publications. http://www.ripublication.com trusted third party broker that evaluates the trust value with an associate trust rating engine. It studies the QoS attributes related with cost and benefit. It assumes that the experience factor as a feedback reputation is as an algorithmic growth curve. This, however does not consider the negative reputation while all ratings are available and valid. In [4], focuses on direct trust implementation established in handling dynamic and real-time nature of the cloud services. To reduce the fake reputation, consumers record their service request and delivery based on the prescribed multiple trust attribute values. The study claims that better real-time performance can be achieved by issuing IOWA operator to establish time series global trust degree. However, it is based on prediction on statistical usage history and past experience of multiple interaction with the CSP. Meanwhile, in [6] proposes a trust framework whereby customers have an ability to give feedback weighting to the CSPs based on the service experiences. It is based on the resource provider capabilities and consumer majority consensus whereby the trust management is capable to differentiate between expert and amateur consumers. It is resulted to the determination of which CSP is trusted and which are not. The trust feedback will be calculated based on the total score by consumers divided by all of the total consumer feedbacks. This will surely make the system vulnerable to cloud service attack such as the Sybil attacks or the whitewashing attacks. However, the researcher provides some counter measures for the false reputation problems. He suggested a counter mechanism where a credibility model is proposed. In this model, the consumers will be divided into two categories called the Expert Cloud Service Consumer (ECSC) and the Amateur Cloud Service Consumer (ACSC). In addition to that, he also proposed that the consumer which had used the cloud service for a longer time would be prioritized in terms of his vote count. This two will be added up to make sure which consumer has the higher vote prioritization in terms of determining the trust level of the cloud network. In a cloud federation, CSPs are competitively offering services. They prompt to offer the below expected QoS value by customers. Based on the reliability and reputation, [5] establishes a formal specification to handle the vast reliability issues. They use agent based to pass the reputation from a node to another based on averaging those recommendations. Recommendation reliability is by means of time-step from the previous value/steps. The false reputation errors made by previous agents will be considered rather than discarded. This means that the false reputations are always being accumulated and averaging at every step.

Figure 1: Trust management framework PROPOSED FRAMEWORK We propose a slightly different framework adopted from [6] for federated cloud infrastructure. We introduce the trust node acting as a broker between customers and CSPs (see Figure 1). This broker node is actually running trust manager services as it is responsible in handling all trust related issues such as creating a session key and also managing the reputation of each CSP. This is also considered that if the associated broker is failed, the consumer service request can be forwarded to other available neighbouring brokers as its trust values are updated occasionally. In a nutshell, the broker acts as the highest trust authority of the federated cloud architecture which is quite similar to the Certificate Authority (CA) in the Web System Architecture. The CSP still has its authority by specifying its services as well as allowing only the trusted another CSP to use the services. The trust manager will assign a session key to each intermediate node which later will transfer the session key to the cloud nodes using Quantum Key Distribution (QKD) protocol for authentication purposed between cloud networks. Previously, Distributed Hash Table(DHT) is used for the authentication purpose but DHT can be duplicated so there are some vulnerabilities to the federated cloud infrastructure. This session key is also used for the CSPs to determine the other cloud providers who are in the same federated cloud. A reputation feedback will be used to determine the trust level of the cloud node. This is done by cloud consumers that will give a reputation based on their experience of cloud services. If the services requested are successful, it will give the CSP a trust value of +1, otherwise the reputation of the CSP will be deducted by 1. Thus, we can consider that, Rt represents the total reputation of the CSPs, Rc is the reputation given by the consumer and k is the total number of cloud consumer that submit a reputation. Thus, the total reputation gained by a CSP is given by 𝑅𝑡 = ∑1𝑘 𝑅𝑐

(1)

In this case 1, it seems fair as the consumer only gives reputation to the CSP that serves their request. Nevertheless, in federated cloud environment, the CSP can pass the request to the other CSP to execute the job on its behalf. The

10602

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 21 (2016) pp. 10601-10605 © Research India Publications. http://www.ripublication.com reliability of trust value might be inaccurate if the other cloud nodes are the one who fail to do the job. If this is the case, the relaying cloud may deduct the reputation of the problematic cloud accordingly. Let say Cr represents the reputation among CSPs. By using the same principle, the owner of the cloud can give the reputation to other related clouds as well. Thus, the cloud reputation can be modelled by 𝑙

𝐶𝑅 = ∑𝑘1 𝑅𝑐𝑝

Algorithm 1. This algorithm will consider consumers’ false reputation, hence its purpose is to mitigate the Sybil attack.

(2)

where Rcp represents the reputation given by CSP and l represents the total amount of clouds that gives reputation to a cloud. So, the real reputation of a CSP can be denoted as 𝑅𝑐𝑟 =

𝑅𝑡 𝐶𝑟 ∗100)( ∗100) 𝑥 𝑙

(

200

*100

(3)

where Rcr is the real cloud reputation for the specific cloud. In this case, we denote cloud reputation in two perspectives where the consumer holds 50% of the total reputation marks while the CSP holds the other 50% of the total reputation marks against the other clouds. Due to the security issues in feedback mechanism such as Sybil attack, where it might have a consumer who forges his identity and purposely gives false reputation to the cloud node. Therefore, a contingency plan to counter this is a must. We can appoint a list of trusted consumers based on the list of consumers that voted for the CSPs. The trusted consumers can be selected based on the number of services they have used on any CSPs which is inside the federated cloud compared to the other consumers who use the federated cloud. Considering that we want to distinguish the trusted consumers who give trust feedback to a CSP, A. In this scenario, we do not determine the number of services used by consumer X , Scx. We determine the average number of services used by all the consumers that give reputation to A, Sau, where n is the number of consumers that vote for cloud A and S v is the services used by other voters. This is can be computed by 𝑆𝑎𝑢 =

(∑𝑛−1 𝑆𝑣 )−𝑆𝑐𝑥 1

(4)

𝑛−1

The other things that we have to consider are to determine if Scx is more than Sau, to determine if consumer, A gives reputation to cloud X is within the value of majority of consumers have given, Rm. For an example, in the case of consumerA assigns a positive reputation, 1 to the cloudX 1, whereby most of the consumers give −1 reputation. Therefore, consumerA can be only trusted, Tu if 𝑇𝑢 = 1 ⇒ 𝑇𝑢 = (𝑆𝑐𝑥 > 𝑆𝑎𝑢 ), (𝑅𝑚 == 1)| (𝑆𝑐𝑥 > 𝑆𝑎𝑢 )&&(

𝑆𝑐𝑥 2

== 𝑅𝑚 ) ∈ 𝑇𝑢

(5)

The algorithm to establish a trust level by a cloud consumer to it is service provider based on experience factor can be seen in

RESULTS AND DISCUSSION As mentioned earlier, the total trust result is actually influenced by the reputation feedback from consumers. It actually reflects either the CSP is reputable or not in handling the consumers’ request. In establishing that, we conduct an experiment with both positive and negative reputations where the trustworthy values are vague. We use the same strategies to collect the data set as in [6] with 10,076 feedbacks from 6,892 consumers in evaluating 113 cloud services. The result can be seen in Figure 2. It shows the final trust value over consumer’s feedback as accumulated in a CSP. The trust value without consumer experience feedback is high as can been seen in the Figure 2. This value is degraded as it

10603

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 21 (2016) pp. 10601-10605 © Research India Publications. http://www.ripublication.com introduces the trust element to the experience feedback factor [6]. It shows that the consumers’ feedbacks will significantly affect the final trust value associated to the CSP. In our case, the proposed objective of the algorithm is to distinguish between the experienced and inexperienced consumers. This will mitigate the Sybil attack problem which happened in clouds, where attackers with multiple identities giving false feedback reputations (high or low value) to the CSPs, hence altering the total reputation for the CSPs significantly. Based on the previous research, the researcher mentioned that trust result produced when using the majority consensus factor (consumer experience factor) must be lower than without the experience factor [6]. We argue that, it does not necessary low but can be high, and the most important is that inexperienced consumers or attackers cannot change or influence the trust value significantly.

To determine whether my hypothesis is true, we have to compare the rate of change of consumers’ experience using proposed algorithm and [6]. This test is performed on three clouds. We use IBM SPSS for the data analysis purposes to show that the proposed approach is robust to the Sybil attacks. Although the trust result will affect the final reputation for a cloud, this will not affect the way we determine the best CSP. Based on the Table 1, we can see that the standard deviation (SD) in [6] is higher than the proposed algorithm. This means that every reputation feedback will give a huge weightage to the final trust value. However, with the same input data, the proposed algorithm shows a lower SD value. This is important in order to handle the Sybil attacks problem whereby our objective is that the trust value cannot be changed so much by the attackers. In this case, when random values are added, even though with a very low or very high value, it won’t significantly reduce or increase the cloud reputation as a whole. Thus, the final value will remain close to the original mean value. Table 1: Descriptive statistic of both algorithms

Figure 2: Consumer experience factor It is because, during Sybil attack, intruders can give any inexact trust value to the CSPs depending on their objectives either to increase or decrease its reputation. Thus, the value is not necessarily has to be lower, but sometimes higher as compared to the value that they are supposed to commit. In our case, we compare the value of the potential attackers’ reputation with some selected experienced consumers and majority value by the total consumers. Because of this, we can no longer take the final trust value as a determination factor. This is because, we want to make sure that when an intruder enters a fake reputation into the CSP, the fake result will not alter the total result of the CSP’s trust as much.

Algorithm

N

Min

Max

Mean

Proposed H.Noor[6] Valid N

100 100 100

6.9 4.5 -

8.32 7.30 -

8.15 6.12 -

Std. Dev 0.29 0.74 -

CONCLUSION We argue that the final value of trust result in reputation feedback mechanism that established in most of the study cannot be used as an absolute solution in reflecting trustworthiness of CSP. It prompts to security attack such as Sybil attack. Instead of the final trust value, we propose a rate of change value that considered robust and can give a significant and meaningful evaluation. However, it must come together with a cross reference to other consumers or CSPs. As a result, whenever there are attacks that try to give false feedback reputation to a CSP, the final trust value does not change drastically. REFERENCES

The reason behind the rate of change has to be considered due to the difficulty for the CSPs to distinguish either the reputation feedback is trustworthy or not. For an example, there is a case of consumer’s reputation being manipulated by others and falsifying the actual values. Compared to [6], a cross reference has been made in the terms of service request time with other consumers. In this case, we double check whether the computation is failed or successful by comparing the average time to complete the requested services with other consumers. Let say the consumers give bad reputation due to failed services or take too long to complete, hence the bad reputation given by the consumers would be considered as true and vice-versa for consumers who give too high reputations compared to majority of the consumers.

[1]

[2]

[3]

[4]

10604

D Bernstein and D Vij. Intercloud exchanges and roots topology and trust blueprint. In Proc. of 11th International Conference on Internet Computing, pages 135–141, 2011. 
 John R. Douceur. The sybil attack. In Revised Papers from the First International Workshop on Peer- toPeer Systems, IPTPS ’01, pages 251–260, London, UK, UK, 2002. Springer-Verlag. Y. Kim and K. G. Doh. Quantitative trust management to support qos-aware service selection in service-oriented environments. In Parallel and Distributed Systems (ICPADS), 2013 International Conference on, pages 504–509, Dec 2013. 
 X. Li and J. Du. Adaptive and attribute-based trust

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 21 (2016) pp. 10601-10605 © Research India Publications. http://www.ripublication.com

[5]

[6]

[7]

model for service level agreement guarantee in cloud computing. IET Information Security, 7(1):39–50, March 2013. 
 F.Messina,G.Pappalardo,D.Rosaci,C.Santoro, and G. M. L. Sarn. A trust model for competitive cloud federations. In Complex, Intelligent and Software Intensive Systems (CISIS), 2014 Eighth International Conference on, pages 469–474, July 2014. 
 Quan Z. Noor, Talal H.and Sheng. Trust as a Service: A Framework for Trust Management in Cloud Environments, pages 314–321. Springer Berlin Heidelberg, Berlin, Heidelberg, 2011. 
 Rajani Sharma and Rajender Kumar Trivedi. Literature review: Cloud computing–security issues, solution and technologies. International Journal of Engineering Research ISSN, pages 2319–6890, 2013. 


10605