credibility-based cloud media resource allocation algorithm

Credibility-based cloud media resource allocation algorithm

Ruichun Tang a,b,n, Yuanzhen Yue a, Xiangqian Ding a, Yue Qiu a

a College of Information Science and Engineering, Ocean University of China, Qingdao 266100, Chinab State Key Laboratory of Digital appliances, Qingdao 266101, China

a r t i c l e i n f o

Article history:Received 16 January 2014Received in revised form15 May 2014Accepted 12 July 2014

Keywords:Cloud mediaContinuous double auctionResource allocationCredibility

a b s t r a c t

Traditional cloud media resource allocation algorithms have the problem of low efficiency duringresources allocation in cloud environment, which is caused by lacking credibility between mediaresource nodes, a credibility-based cloud media resource allocation (CCMRA) algorithm is proposed inthis paper. According to the continuous double auction mechanism, the resource applicants and resourceowners submit their requests to the allocation agents. Based on the total credibility, the allocation agentsallocate the media resources to get the optimal allocation sequence for higher allocation efficiency andQuality of Service (QoS). Finally, the effectiveness of the proposed algorithm is proved by the simulation.

& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

In cloud environment and wireless network, there exist a lot ofdynamic and heterogeneous media resources, and these mediaresources are shared by multiple resource applicants simulta-neously (Wang et al., 2011). Allocating media resources efficientlycan not only improve the media resource utilization greatly, butalso increase the economic QoS and performance QoS (Zhanget al., 2013). How to improve the allocation utility is an eagerproblem to be solved nowadays.

Researchers have studied a lot about the utility problem ofmedia resource allocation. Among them, the ant colony algorithm,particle algorithm, genetic algorithm etc. are representative anduniverse (Xing-wei et al., 2012). The particle swarm algorithm hasbeen proposed in Gong et al. (2012) to solve the problems ofresource scheduling and maximize the workload profits. Thedouble auction based Nash equilibrium algorithm has been pro-posed in Sun et al. (2010) to allocate media resources, and mean-while it takes the social and economic QoS into account.

All the above papers have solved the utility problem of mediaresource allocation, but none of them considered the issue ofmedia resources credibility, which resulted in the low resourceallocation efficiency. In cloud environment, the already allocatedmedia resources may fail to arrive at the resource applicantsbecause of the network instability or the nodes dishonesty,

otherwise, the already applied resource applicants may not needthe resources because the tasks have been finished or the price hasbeen changed. What is more, the credibility of one of the transac-tion participants affects the other’s utility directly.

As a result, some researchers put forward the concept ofresource credibility. Concept of node trust based on double auctionhas been proposed in Shi-Wei and Yu (2011), it combined the nodecredibility with the resource price which had improved thesuccess rate and stability of resource allocation to some extent.The credibility was defined as an attribute of quality of informa-tion in Ciftcioglu and Yener (2012), it considered the networkutility which depended on both information credibility and time-liness to find the optimal power allocation. Trust model in optimalresource allocation for a virtual organization has been proposed inShu-gang and Jian-hua (2011) by using trust mechanism toimprove service satisfaction, the resource trust was divided intodirect trust and indirect trust, then resources were allocated basedon the total trust value of each resource allocation chain. All theabove papers considered the credibility of resources, but ignoredthe credibility between the resource nodes.

In this paper, the concept of credibility between the resourcenodes is proposed considering both the resource owners and theresource applicants. Then we set the maximum utility function toget the optimal allocation sequence. Finally the allocation agentsallocate the media resources based on the credibility of theoptimal allocation sequence, which can avoid the resource wastecaused by network instability.

This paper is organized as follows. Section 1 is the researchenvironment in cloud. Section 2 formulates the cloud mediaresource allocation model. Section 3 presents the resource pricing

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jnca

Journal of Network and Computer Applications

http://dx.doi.org/10.1016/j.jnca.2014.07.0181084-8045/& 2014 Elsevier Ltd. All rights reserved.

n Corresponding author at: College of information Science and Engineering,Ocean University of China, Qingdao 266100, China. Tel.:þ86 131 5320 6505.

E-mail address: [email protected] (R. Tang).

Please cite this article as: Tang R, et al. Credibility-based cloud media resource allocation algorithm. Journal of Network and ComputerApplications (2014), http://dx.doi.org/10.1016/j.jnca.2014.07.018i

Journal of Network and Computer Applications ∎ (∎∎∎∎) ∎∎∎–∎∎∎

www.sciencedirect.com/science/journal/10848045

www.elsevier.com/locate/jnca

http://dx.doi.org/10.1016/j.jnca.2014.07.018



mailto:[email protected]





strategies. The media resource credibility is proposed in Section 4.Section 5 is the CCMRA algorithm. Finally the simulation is carriedout in Section 6 and Section 7 is the conclusion.

2. Research environment

As shown in Fig. 1, the resource applicants, resource ownersand resource agents are all in the cloud environment. Firstly, theresource owners and the resource applicants submit their owninformation (e.g., the task deadline, the number of requiredresources etc.) to the resource agents simultaneously. Then, theresource agents get the optimal allocation sequence based on theCCMRA algorithm. Finally, the resource owners and resourceapplicants form transactions to allocate resources based on theoptimal allocation sequence.

When a resource applicant needs to apply for media resources,it sends the number of required resources, the deadline of the taskand other attributes (e.g., the budget and the highest price withinits affordability etc.) to the allocation agents. The allocation agentsallocate resources based on the resource applicants’ highest priceand the resource owners’ lowest price. Then the resource agentsallocate resources as the maximum utility sequence based on themedia resource credibility.

3. Cloud media resources allocation model

In the network, media resource allocation is a many-to-manytransaction, especially in the cloud environment which is of morecomplexity and variability. As shown in Fig. 2, the service layer, theuser layer and the agent layer of the CCMRA model correspond tothe sellers, buyers and auctioneers of the CDA model respectively.The resource applicants and resource owners allocate mediaresources through the allocation agents. Resource applicants sub-mit the number of required resources and the price to theallocation agents. The resource owners uninterrupted submitthe number of remaining resources and acceptable price to theallocation agents. After receiving the information of the resourceapplicants and resource owners, the allocation agents get theallocation sequence with the maximum utility value consideringthe price, deadline and the credibility based on the CCMRA, thenallocate the media resources to the corresponding resourceapplicants.

Definition 1. Let U ¼ fu1;u2;…;umg be the set composed of mresource applicants, each task of resource applicant ui is ti, so the

task set of U can be described as T ¼ ft1; t2;…; tmg. And ti has fourattributes ti ¼ ftidi; li; bi; dig; iA ½1;m�, where tidi is the ith task’sidentify, li is the ith task’s length, bi is the ith task’s budget, and diis the deadline of the task.

Definition 2. Let O¼ fo1; o2;…; ong be the set composed of nresource owners, each resource of resource owner oj is rj, so theresource set of O can be described as R¼ fr1; r2;…; rng. And rj hasfive attributes rj ¼ fridj; cpuj; stj; lpj;hpjg; jA ½1;n�, where ridj is thejth media resource’s identify, cpuj is the jth media resource’scomputing ability of solving the task, stj is the start time to dealwith a new task (i.e. the current workload of resource rj), lpj is thejth media resource’s lowest price, and hpj is the jth mediaresource’s highest price.

Definition 3. The media resources allocation probability matrix asfollows:

P ¼

p11; p12;…; p1np21; p22;…; p2n: : :

pm1; pm2;…; pmn

0BBBB@

1CCCCA:

where pij is the probability of resource applicant ui submitting atask to resource rj, and

0rpijr1 and ∑m

i ¼ 1pij ¼ 1 and ∑

n

j ¼ 1pij ¼ 1;

which means that the probability should be at the range of [0,1],and the total sum of each row or column should be 1.

4. The media resource allocation participants’ pricingstrategies

During the resources allocating, the resource allocation parti-cipants (e.g., resource applicants and resource owners) havedifferent pricing strategies, and submit their price and require-ment to the allocation agents. The purpose of allocation agents isto achieve the economic QoS and performance QoS to maximizethe social benefits.

4.1. The resource applicants’ pricing strategy

As described in Anthony and Jennings (2003), there are manyfactors which may affect the resource applicants’ price, amongthem the number of remaining resources and average remainingFig. 1. Cloud media resources allocation research environment.

Fig. 2. Credibility-based media resource allocation model.

R. Tang et al. / Journal of Network and Computer Applications ∎ (∎∎∎∎) ∎∎∎–∎∎∎2





time are the two main factors that drive the resource applicant’sbehavior in making the pricing strategy. So in this paper, we onlytake the two factors into consideration.

The time constraint is described by (1), which means resourcerj can be allocated to resource applicant ui is on condition that themedia resource can finish the task within the deadline.

di�stj� li=cpujZ0 ð1Þ

The price constraint is described by (2), which means the priceof resource should not be higher than the resource applicants’price.

bi=liZ lpj ð2Þ

It is meaningful to consider these two following factors onlywhen task ti and resource rj meet time constraint in (1) and priceconstraint in (2).

4.1.1. Applicants’ price considering the remaining media resourcesIn order to protect its own interests, the resource applicants’

price should be at least lp, which is the average price of remainingresources, and lp ¼ 1=n

� �∑n

1lpj. With the decrease of remainingresources, the resource applicants’ price should be higher in orderto purchase the limited resources. That ensures the limited mediaresources can be applied by more urgent resource applicants toimprove the media resource utilization. So the price consideringremaining media resources is described as follows:

bidresourcei ðtÞ ¼ lpbi=li

þ 1� lpbi=li

!1� nt

i

nmaxi

� �1=α !bili

¼ lpþ bili� lp

� �1� nt

i

nmaxi

� �1=α

ð3Þ

where nti is the number of remaining media resources that can be

applied by task ti at time t, and nmaxi is the maximum number of

media resources that can be applied by task ti. Different applicant’sprice curve can be adjusted by changing αðαA ð0:01;100ÞÞ. Whenαo1, the resource applicants maintain a low price value until thenumber of the remaining resources gets close to zero. When α41,the resource applicants maintain a high price value close to bi=li.As shown in Fig. 3, we set lp¼50, bi=li¼65, and nmax

i ¼100, so therange of nt

i is from 0 to 100. We can make the conclusion that theresource applicants’ price decreases when the number of remain-ing resources increases, and different curves can be adjusted bychanging α.

4.1.2. Applicants’ price considering the average remaining timeThe remaining time rtijðtÞ of task ti to resource rj is

di�stj� li=cpuj, and the average remaining time of applicant ui isdescribed as follows:

rtiðtÞ ¼ ∑n

j ¼ 1ðrtijðtÞωijÞ=nmax

i ð4Þ

where

ωij ¼1 if rtijðtÞZ00 otherwise

�

With the decrease of average remaining time, the applicants’price should be higher. That ensures the resource applicants canapply for enough media resources to finish the task within dead-line to improve the media resource utilization. So the priceconsidering the average remaining time is described as follows:

bidtimei ðtÞ ¼ lpþ bi

li� lp

� �1�rtiðtÞ

rtmaxi

� �1=β

ð5Þ

where rtmaxi is the ith resource applicants’ maximum remaining

time, and 0:01rβr100, β plays the same role as α in (3) and isused for controlling convexity degree of the curve.

From what has been discussed above, the resource applicants’pricing formula considering both the remaining resources and theaverage remaining time is described as follows:

bidiðtÞ ¼ α0bidresourcei ðtÞþβ0bidtimei ðtÞ;

0rα0; β0r1 ð6Þwhere α0;β0 are the weights of the remaining resources and theaverage remaining time respectively to regulate the effectivenessof these two factors, and α0 þβ0 ¼ 1. α0 ¼ 1 means that only theremaining resources is considered, β0 ¼ 1 means that only theremaining time is considered, and α0;β0Að0;1Þ means that bothconstraints are taken into account. When time is enough we setα04β0, on the other hand, when resources are rich we set α0oβ0.

4.2. The resource owners’ pricing strategy

The resource owner aims at maximizing its benefit. For thispurpose, it tries to sell its media resources at a higher price andcompetes with other resource owners for accepting more tasks.We assume that the start time of resource stj is zero initially andthe resource owner sets its price to lowest price lpj for accepting atask. After accepting a task the media resource updates its starttime stj and sets its price to the maximum price hpj. Gradually, thestart time stj is decreased and gets close to zero, which means thatthe media resource will be released soon and can be applied byother resource applicants. By decreasing the start time stj, theresource owner decreases its media resource price, and in the casethe start time stj ¼ 0, it sets the price to the lowest price lpj toensure that it can be easily applied by other resource applicants,which can improve media resource utilization. The resource ownerdetermines its media resource price by (7) and submits the resultto the allocation agents.

rpjðtÞ ¼ lpjþðhpj� lpjÞstjðtÞwljðtÞ

� �1=σ

ð7Þ

where rpjðtÞ is the price of media resource rj at time t, stjðtÞ is thecurrent workload or start time of a new task at time t, and wljðtÞ isthe workload of media resource rj after the last allocation.Gradually stjðtÞ decreases, the current task is going to be finishedand the media resource will be released soon, then the price of themedia resource rpjðtÞ decreases in order to be applied by otherresource applicants, which can improve the media resourceutilization. σ plays the same role as α;β in (3) and (5) respectively,Fig. 3. Resource applicants’ price considering remaining resources.

R. Tang et al. / Journal of Network and Computer Applications ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3





and is used for controlling convexity degrees of the curve. Whenσo1, the media resource maintains a low price until the mediaresource workload gets close to wl. When σ41, the mediaresource maintains a high price close to hp. As shown in Fig. 4,we set lpj¼45, hpj¼80, and wljðtÞ¼100, so the range of stjðtÞ isfrom 0 to 100. The media resource price rpjðtÞ increases with themedia resource workload stjðtÞ, and different curves can beadjusted by changing the value of σ.

4.3. The allocation agents’ pricing strategy

The resource applicants and resource owners submit their priceand requirement to the allocation agents, the allocation agentssort the resource applicants’ prices bidiðtÞ in descending order andmedia resource prices rpjðtÞ in increasing order (Izakian et al.,2009). If the highest bidiðtÞ is more than or equal to the lowestrpjðtÞ, then the final price is described as follows:

f pðtÞ ¼ 12ðbidmax

i ðtÞþrpminj ðtÞÞ ð8Þ

This section describes the pricing strategies of the mediaresources and the resource applicants respectively, and gives thefinal price between them. Traditional resource allocation algo-rithms will allocate resources according to these prices, but werecommend the concept of media resource credibility in thispaper, as in Section 4. The success rate of resource allocation canbe improved obviously by considering credibility, which canimprove the media resources utilization simultaneously.

5. Media resource credibility

In cloud environment, the credibility is the trustworthy ratebetween different resource nodes, which is changing with timeand network. The successful probability will be low if the cred-ibility between them is low. So we should avoid allocatingresources to the nodes with low credibility. To solve this problem,we take the nodes credibility into consideration during theresource allocation in this paper. The failure probability is lowerwhen the credibility is higher, which results in the higherresources utilization.

A transaction is formed when the resource applicant’s pricebidiðtÞ is not lower than the media resource price rpjðtÞ. Thetransaction is defined as ti;jðf pðtÞÞ when the transaction price isf pðtÞ and the result of a transaction is either failure or success.

We denote the outcome for the ith resource applicant as eui Af0;1gand for the jth resource as erj Af0;1g, where 0 represents failureand 1 represents success. The success probability of a resourceapplicant (i.e. Pðeui ¼ 1Þ) is denoted as pðuiÞ and that of the mediaresource (i.e. Pðerj ¼ 1Þ) is denoted as pðrjÞ. For example, aftertransaction ti;jðf pðtÞÞ is formed, if eui ¼ 1 while erj ¼ 0, that meansresource applicant ui has paid for a resource, but rj did not providethe resource.

In cloud environment, the credibility of one of the transactionparticipants affects the other’s utility directly. If resource applicantui has a transaction with media resource rj, first the utility functionof resource applicant ui is described in (9), the utility of resourceapplicant ui ¼ bi=li� f pðtÞ when media resource rj is trustworthy,and the utility of resource applicant ui ¼ � f pðtÞ when mediaresource rj is dishonesty, then we can get the expected utilityfunction of resource applicant ui as in (11). On the other hand, theutility function of media resource rj is described in (10), with theconsideration of whether the resource applicant is trustworthy, wecan get the expected utility function of media resource rj as in (12).

The utility functions of resource applicant and media resourceare described as follows:

uui ðti;jðf pðtÞÞ; erj Þ ¼

bi=li� f pðtÞ; erj ¼ 1

� f pðtÞ; erj ¼ 0

(ð9Þ

urj ðti;jðf pðtÞÞ; eui Þ ¼

f pðtÞ� lpj; eui ¼ 1

� lpj; eui ¼ 0

(ð10Þ

The expected utility functions of resource applicant and mediaresource are described as follows:

uui ðtÞ ¼ uu

i ðt;1ÞpðrjÞþuui ðt;0Þð1�pðrjÞÞ

¼ bi=lipðrjÞ� f pðtÞ ð11Þ

urj ðtÞ ¼ ur

j ðt;1ÞpðuiÞþurj ðt;0Þð1�pðuiÞÞ

¼ f pðtÞpðuiÞ� lpj ð12Þ

To a conclusion, the total utility based on the credibility of boththe resource applicants and the media resources is described asfollows:

Utiðf pðtÞÞ ¼ uui ðtÞþur

j ðtÞ¼ bi=lipðrjÞ� f pðtÞþ f pðtÞpðuiÞ� lpj ð13Þ

The higher utility value calculated by (13) means the transac-tion of resource applicant uiand media resource rjmore trust-worthy and the allocation success rate higher. As whenUtiðf pðtÞÞo0, we can be sure that the media resource nodes iand j have very low credibility. So if Utiðf pðtÞÞo0 we can assumethat the transaction between i and j is failed. By consideringcredibility, the problem of failure allocation caused by networkinstability and nodes dishonesty can be avoided to some extent.Section 5 gives the algorithm of how to get the optimal allocationsequence.

6. The CCMRA algorithm

As a non-cooperative game (Grosu and Chronopoulos, 2005),the objective of cloud media resources allocation is to get highersuccess rate and each resource applicant or resource ownertries to maximize its own performance-QoS and economic-QoSindependently.

In the cloud media resource allocation system, we assume thatonce the task distribution of the resource applicants is chosen, itwill be consistent throughout the system. Our goal is to get theoptimal allocation for maximum media resource utility on thebasis of considering the media resource credibility.

Fig. 4. Request value based on workload.






Definition 4. Optimal allocation sequence. We assume that

Ω¼ pijUtiðf pijÞ

¼

p11Utiðf p11Þ;…; p1nUtiðf p1nÞp21Utiðf p21Þ;…; p2nUtiðf p2nÞ: : : : :

pm1Utiðf pm1Þ;…; pmnUtiðf pmnÞ

0BBBB@

1CCCCA

is the set of the entire possible allocation matrix. Γ ¼ fgð1; j1Þ;ð2; j2Þ;…; ðm; jkÞ; jkA ½1;n� is all the possible sequence while jk isdifferent with each other, and Γn ¼ fð1; jn1Þ; ð2; jn2Þ;…; ðm; jnkÞg; jkA½1;n� is the sequence with the max utility as shown in (14). Theutility function is shown in (15):

UtiðΓnÞ ¼ max UtiðΓÞ ð14Þ

UtiðΓÞ ¼ ∑ði;jÞAΓ

pijUtiðf pðtÞÞ ð15Þ

According to the analysis above, the core part of CCMRAalgorithm is described as follows.

The CCMRA Algorithm

1. Initialize R, U, Pm�n.2. For each jA ½1;n� do3. Update rj ¼ ðridj; cpuj; stj; lpj;hpjÞ to the allocation agents.4. End for5. For each iA ½1;m� do6. Submit ti ¼ ðtidi; li;bi; diÞ to the allocation agents.7. End for8. For each iA ½1;m� do9. Calculate bidiðtÞ by (6).

10. End for11. For each jA ½1;n� do12. Calculate rpjðtÞ by(7)13. End for14. Quick sort ðbid1ðtÞ; bid2ðtÞ;…; bidmðtÞÞ into

ðbid01ðtÞ; bid02ðtÞ;…; bid0mðtÞÞ by decreasing order. Quick sort ðrp1ðtÞ; rp2ðtÞ;…; rpnðtÞÞ into ðrp01ðtÞ; rp02ðtÞ;…; rp0nðtÞÞ byincreasing order.

15. If minðbid0iðtÞÞr maxðrp0jðtÞÞ do16. Calculate f pðtÞ by (8).17. While not max credibility utility do18. For each Γ ¼ ðð1; j1Þ; ð2; j2Þ;…; ðm; jkÞÞ do19. Calculate UtiðΓÞ by (13).20. If max UtiðΓÞoUtiðΓÞ,then21. max UtiðΓÞ ¼UtiðΓÞ22. End for23. End while24. UtiðΓnÞ ¼ max UtiðΓÞ25. Output the optimal allocation sequence with the max cred-

ibility utility Γn ¼ fð1; j1nÞ; ð2; j2nÞ;…; ðm; jknÞg.

26. End

The inputs of this algorithm are the media resource set R, theresource applicant set U, and the probability matrix Pm�n duringone period of transaction. The output is the optimal allocationsequence with the maximum credibility utility. The resourceapplicants and media resource owners determine their price andrequests and submit themselves to the allocation agents from step2 to step 7, and then the allocation agents preprocess the mediaresource owners and the resource applicants values from step 8 tostep 16. From step 17 to step 22, all the possible solutions arecompared to make sure which one is the optimum efficientallocation, and the comparison is a process of depth-first searchso as to improve the performance of the proposed algorithm. Theprocess of CCMRA can be described as in Fig. 5.

7. Simulation and performance evaluation

In order to evaluate the performance of the algorithm pre-sented in this paper, we implement this algorithm by the Cloud-Sim toolkit (Calheiros et al., 2011). In this paper, the allocationagents provide the resource applicants with the cloud mediaresources as the optimum allocation sequence.

As shown in Table 1, we set 200–900 tasks to apply for thecloud media resources. Each task is submitted according to Poissondistribution after its previous tasks, the length of each task isconsidered as a random number within [100 000,200 000], thedeadline di of task ti is set according to (16), and the budget bi oftask ti is set according to (17).

di ¼ stjþrandomli

1:1ncpuj;

li0:9ncpuj

� �ð16Þ

bi ¼ li Urandomð0:9lp;1:1hpÞ ð17Þwhere lp and hp are the mean values of the media resources’ lpand hp respectively.

Moreover, we set 30–50 cloud media resources, the computingability cpuj is within [1000,2000], the start time stj of each mediaresource is set 0 initially and is updated after each transaction. Thelowest price lpj is within [100,500], and the highest price hpj iswithin [1000,1400].

Fig. 5. The process of CCMRA.






After submitting each task, the allocation agents are activatedand try to allocate this task by bidding for resource owners. Andthe task is deleted when it is finished. For evaluating theperformance-QoS and the economic-QoS, the FIFO algorithm(Chen et al., 2013), the CDA algorithm (Sun et al., 2010), theRandom Algorithm (RA) (Touray et al., 2012) and the GreedyAlgorithm (GA) (Zhang and Zhu, 2013) are used to compare withCCMRA algorithm in the successful execution rate and theresponse time respectively.

7.1. Successful execution rate

The successful task means that the task completely accom-plished within its deadline, and the successful execution rate ofthese m tasks is described as follows:

rate¼∑mi ¼ 1θi

mð18Þ

where

θi ¼1 Dirdi && bidiZrpj0 others

�

where Di is the finish time of task ti, and the constraint Dirdimeans the task ti can be finished before deadline. The constraintbidiZrpj means that the cloud media resource rj can be allocatedto task ti. Task ti is successful if it meets both the time constraintand price constraint.

As shown in Figs. 6 and 7, when the task number is small, thesuccessful execution rates of these five algorithms are high andhave little difference. However, with the task number increasing,the successful execution rates of FIFO, CDA, RA and GA decreasequickly while the CCMRA decreases slowly and maintains a highlevel. The price constraint, the time constraint and the credibilityare all taken into consideration in CCMRA, so the CCMRA hasbetter performance-QoS than the other four algorithms when thetask number is large, which is more suitable for the cloudenvironment with large amounts of tasks.

7.2. The response time

The response time of the media resources can evaluate theperformance of the cloud media resources allocation. The responsetime is described as follows:

Time¼ ∑n

j ¼ 1timej ð19Þ

where timej is the response time of the jth resource, and the tasknumber increases from 200 to 900. The simulation results areshown in Figs. 8 and 9.

Figures 8 and 9 show that the response times of these fivealgorithms are basically same at the beginning as the cloud mediaresources are enough. However, with the increasing number of thetask, the CCMRA algorithm shows better stability and its responsetime is less than the other four algorithms, which means that theCCMRA algorithm has better economic-QoS than the other fouralgorithms when the task number is large.

8. Conclusion

Based on the continuous double auction mechanism, we haveput forward the credibility between the media resource allocationparticipants. On considering the credibility between the media

Table 1

The simulation parameter settings

Tasks Resources

Number Length Number Computing Speed Lowest Price Highest Price

Value

Range[200,900]

[100000,

200000][30,50] [1000,2000] [100,500] [1000,1400]

Fig. 6. Comparison of successful execution rate between the algorithms of FIFO,CDA, and CCMRA.

Fig. 7. Comparison of successful execution rate between the algorithms of RA, GA,and CCMRA.

Fig. 8. Comparison of response time between the algorithms of FIFO, CDA,and CCMRA.

Fig. 9. Comparison of response time between the algorithms of GA, RA,and CCMRA.






resources and media applicants, which can avoid failure caused bynetwork instability and nodes dishonesty, the successful executionrate has been improved obviously. Experimental results haveillustrated that the CCMRA algorithm is efficient for both theresource owners and resource applicants, and has greatlyimproved the media resource utilization and performance-QoS.In addition, the algorithm in this paper is more suitable for cloudenvironment in the area of cloud media resource allocation.

Acknowledgment

This work was supported by the National Technology Sup-ported Project “Community Living Services Integration Solutionsand Standards Research” under Grant no. 2012BAH15F01.

References

Wang X, Chin KS, Yin H. Design of optimal double auction mechanism with multi-objectives. Expert Syst Appl 2011;38(11):13749–56.

Zhang, B, Zhao, Y, Wang, R, 2013. A resource allocation algorithm based on mediatask QoS in cloud computing. In: Proceedings of the 4th IEEE InternationalConference on Software Engineering and Service Science (ICSESS), Beijing,p. 841–4.

Xing-wei W, Xue-yi W, Min H. A resource allocation method based on the limitedEnglish combinatorial auction under cloud computing environment. In: Pro-ceedings of the 2012 9th international conference on Fuzzy Systems andKnowledge Discovery (FSKD). IEEE; 2012. p. 905–9.

Gong YJ, Zhang J, Chung HS, et al. An efficient resource allocation scheme usingparticle swarm optimization. Evol. Comput. IEEE Trans. 2012;16(6):801–16.

Sun D, Chang G, Wang C, et al. Efficient Nash equilibrium based cloud resourceallocation by using a continuous double auction. In: Proceedings of the 2010International Conference on Computer Design and Applications (ICCDA), vol. 1.IEEE; 2010. p. V1-94–V1-99.

Shi-Wei C, Yu P. Credibility-based dynamic resource distribution strategy undercloud computing environment. Comput Eng 2011;11:018.

Ciftcioglu EN, Yener A. Maximizing credibility-based network utility via powerallocation. In: Proceedings of the 2012 IEEE international conference onPervasive Computing and Communications Workshops (PERCOM Workshops).IEEE; 2012. p. 8–13.

Shu-gang M, Jian-hua Y. Research on trust model in optimal resource allocation fora virtual organization. In: Proceedings of the 2011 International Conference onE-Business and E-Government (ICEE). IEEE; 2011. p. 1–4.

Anthony P, Jennings NR. Developing a bidding agent for multiple heterogeneousauctions. ACM Trans Internet Technol 2003;3(3):185–217.

Izakian H, Ladani BT, Zamanifar K, et al. A continuous double auction method forresource allocation in computational grids. In: Proceedings of IEEE symposiumon Computational Intelligence in Scheduling, 2009 (CI-Sched'09), IEEE; 2009. p.29–35.

Grosu D, Chronopoulos AT. Noncooperative load balancing in distributed systems.J Parallel Distrib Comput 2005;65(9):1022–34.

Calheiros RN, Ranjan R, Beloglazov A, et al. CloudSim: a toolkit for modeling andsimulation of cloud computing environments and evaluation of resourceprovisioning algorithms. Softw: Pract Exp 2011;41(1):23–50.

Chen, Bi Yu, et al. 2013. Shortest path finding problem in stochastic time-dependentroad networks with stochastic first-in-first-out property. In: Proceedings of theIntelligent Transportation Systems, IEEE Transactions 14 vol. 4, 1-11.

Touray B, Shim J, Johnson P. Biased random algorithm for load balancing in WirelessSensor Networks (BRALB). In: Proceedings of the 2012 15th International PowerElectronics and Motion Control Conference (EPE/PEMC). IEEE; 2012. p. LS4e.1-1–LS4e.1-5.

Zhang M, Zhu Y. An enhanced greedy resource allocation algorithm for localizedSC-FDMA systems. IEEE Commun Lett 2013;17(7):1479–82.



http://refhub.elsevier.com/S1084-8045(14)00159-3/sbref1


















credibility-based cloud media resource allocation algorithm

Documents