, pp.215-219 http://dx.do.org/10.14257/astl.2016.123.40 Applcaton of Ant colony Algorthm n Cloud Resource Schedulng Based on Three Constrant Condtons Yang Zhaofeng, Fan Awan Computer School, Pngdngshan Unversty, Pngdngshan, 467002 Henan provnce, Chna {Yang Zhaofeng}pdsncyyang@163.com Abstract. Amng at the problem of resource schedulng n cloud computng, an ant colony algorthm based on the three constrant condtons s proposed. The mplementaton of ths method s dvded nto three steps: Wth the help of the nformaton elctaton factor and expect nspraton factor, wth the help of the pheromone ntensty desgn pheromone update strategy, wth the help of three constrants to complete the optmzaton process. Expermental results show that the three constrant condtons ant colony schedulng algorthm can complete the resource schedulng n a shorter tme, and the load dstrbuton of the schedulng results s more balanced. Keywords: cloud computng, resource schedulng, ant colony algorthm, pheromone. 1 Introducton At present, there are more researches on user job schedulng of and resource schedulng n cloud computng. For the cloud of job schedulng, Guo proposed the Meta-schedulng model and the Meta-schedulng polcy of a cloud envronment. The model and job schedulng strategy can ensure that the entre data center's power consumpton or carbon doxde emssons mnmum [1]. Alrokayan uses the cloud to expand the cluster to deal wth the local cluster resource scarcty, and use some job schedulng strategy to evaluate the performance of the job, analyze the resource cost. The expermental results show that, n the case of heavy load, the Nave schedulng strategy to use the resources of the cloud wll generate a lot of cost, whle the use of request backfllng and redrecton strategy based on the expanson factor to use the resources of the cloud, the cost s relatvely low [2]. For the resource schedulng n the cloud, Pop proposed a resource load balancng schedulng algorthm based on trust drven. The algorthm consders the trust requrements of user tasks, takes the resource load balancng as the goal, and takes nto account the tme span of the task executon tme and other factors. Smulaton results show that the proposed algorthm can mprove the load balance and reduce the relatve executon tme of the tasks [3]. Beghdad proposed the ant colony optmzaton algorthm, and s used to solve the ISSN: 2287-1233 ASTL Copyrght 2016 SERSC
problem of the classcal travelng salesman problem n computer [4]. Subsequently, a large number of mproved ant colony optmzaton algorthm began to appear, and was appled to the job-shop problem, network routng problem, vehcle routng problem and other problems [5]. At present, the convergence of the algorthm s not hgh. Takng nto account the cloud computng envronment and the grd envronment has a bg dfference, the sze of the cloud s larger, and the computng power of each node s slghtly lower than that of the grd ste resources [6]. However, the servce qualty s not lower than the grd computng, so the effcent resource schedulng algorthm s the key to mprove the performance of the cloud computng [7]. 2 Ant colony algorthm wth three constrant condtons Ant colony optmzaton algorthm s a knd of swarm ntellgence algorthm, whch smulates the process of ant foragng [8, 9]. In ths paper, the update strategy of pheromone s adopted as follows: R ( ) D j t l 0 across(, j) not across(, j) In the formula, R represents the pheromone ntensty, whch affects the convergence rate of the algorthm to a certan extent, and D ndcates the total length of the path of the ant l n the cycle. When lookng for a sutable computng node, frst, t starts from the node to calculate ts own resources. If the remanng resources are suffcent to meet the amount of jobs submtted by the user, then prorty n the allocaton of ther own resources. If the resource s not enough to calculate the mnmum amount of resources allocated to the user, then accordng to the search algorthm starts searchng for other sutable cloud computng resources. The Ant colony optmzaton allocaton algorthm proposed n ths paper s realzed n ths part. In order to avod stagnaton search, the search wll be carred out wthn a certan range, the purpose s to reduce the algorthm tself brngs network overhead. If there s stll no approprate resources n all of the sub nodes, then the node report to ask for nstructons to the man schedulng node removed user data mrrorng the nodes n the cluster partton. The slave node doman as an undrected graph G(V, E),In whch V s the set of all slave nodes n regonal Area, E s the network set whch connecte to each slave node. In the cloud computng network, t s dvded nto several sub regons, and then the same number of ants are assgned to each regon. Each group of ants search only n ther own regon to fnd a sutable computng node, that s, n the E to fnd an optmal path. The followng parameters should be consdered n the metrc: (1) Expect the executon tme, ET ( f ) refers to the path of f at the end of the computng resources to cope wth ths job takes tme; l (1) 216 Copyrght 2016 SERSC
(2) Network latency: ND ( f ) refers to the maxmum network delay generated by path f; (3) Network bandwdth: BW ( f ) refers to the maxmum bandwdth provded by the path f. Assumng the feature set of a vrtual machne resource VM s: V v, v, v } (2) { 1 2 3 In the formula, v 1 s the CPU feature, x 2 s the memory feature, x3 s the bandwdth feature. User demand for the dversty of cloud computng resources and preferences, how to make the QoS guarantee. QoS descrpton of the task can usually be used to complete the task of tme, memory, network bandwdth and other parameters to quantfy the QoS. For example, task completon tme QoS as the evaluaton crtera, t nclude the start tme of the task, the completon tme, the end tme, etc., can be selected to complete the task of all tme as the evaluaton ndex. Usually the general expectaton vector of the class task can be descrbed as: X x, x, x } (3) { 1 2 3 In the formula, v 1 s the CPU feature, x 2 s the memory feature, x 3 s the bandwdth feature and at the same tme to meet: j1 The constrant functon for resource selecton s: ET ( f ) ND( f ) Y( x) BW ( f ) 3 x 1. j (4) ET ( f ) T1 x s. t. BW ( f ) T ND( f ) T3 2 (5) Ultmately, the process of determnng the target resource and path s the process of fndng the shortest path and the path and the resource of Y (x).,, are the weght of three constrant condtons. T 1, T 2, T 3 are the boundary constrants, whch meet the QoS constrants n cloud computng envronment. Copyrght 2016 SERSC 217
3 Experment result and analyss CloudSm provdes a data center based vrtual technology, modelng and smulaton capabltes and resource montorng, host to vrtual machne mappng capabltes. The confguraton of the experment as follows: the man frequency 2.0GHz of the dual core CPU, the sze of the 8GB memory, 500GB hard drve, Wndows7.0 operatng system. Accordng to the ant colony algorthm and schedulng work flow, n ths experment, we set the number of tasks from 40 to 200, and the number of nodes s 8. In order to show the dfference, we have to set the node's QoS property gap larger, manly ncludng CPU, memory and network bandwdth. At the same tme, select the tradtonal ant colony algorthm and three constraned condtons ant colony algorthm s proposed n ths paper, n order to form a comparson on the expermental results. The two algorthms are executed 5 tmes and the average value s taken. The tme spent by the task executon s shown n Fgure 1. 800 700 600 Ant colony algorthm proposed n ths paper Classcal ant colony algorthm Tme(ms) 500 400 300 200 100 40 80 120 160 200 The number of tasks Fg. 1. Comparson of the executon tme of the two algorthms From the graph, we can see that the three constrant condtons of ant colony algorthm, ts executon tme s sgnfcantly lower than the tradtonal ant colony algorthm n the mplementaton of cloud computng resource schedulng. In addton, wth the ncrease of the amount of the task, the advantage of ths knd of speed s more obvous. 4 Conclusons In ths paper, based on the ant colony algorthm, a knd of ant colony algorthm based on three constrant condtons s constructed, whch s used for resource schedulng n cloud computng. Expermental results show that the proposed method has faster 218 Copyrght 2016 SERSC
executon speed than the tradtonal ant colony algorthm, and the load balancng of the results s more satsfactory. References 1. Paol, R., Fernández-Luque, F. J., Doménech, G., Martínez, F., Zapata, J., Ruz, R.: A system for ubqutous fall montorng at home va a wreless sensor network and a wearable mote. Expert Systems wth Applcatons. 39, 5566-5575 (2012) 2. Guo, F., Yu, L., Tan, S., Yu, J.: A workflow task schedulng algorthm based on the resources fuzzy clusterng n cloud computng envronment. Internatonal Journal of Communcaton Systems, 28, 1053-1067 (2015) 3. Moschaks, I.A., Karatza, H.D.: Evaluaton of gang schedulng performance and cost n a cloud computng system. Journal of Supercomputng, 59, 975-992 (2012) 4. Alrokayan, M., Vahd, D.A., Rajkumar. B.: SLA-aware provsonng and schedulng of cloud resources for bg data analytcs. 2014 IEEE Internatonal Conference on Cloud Computng n Emergng Markets, pp.665-673(2015) 5. Al-Ameen, M. N., Hasan, Md. R.: The mechansms to decde on cachng a packet on ts way of transmsson to a faulty node n wreless sensor networks based on the analytcal models and mathematcal evaluatons. Proceedngs of the 3rd Internatonal Conference on Sensng Technology, ICST 2008, pp.336-341 (2008) 6. Ahuja, R., De, A., Gabran, G.: SLA based scheduler for cloud for storage & computatonal servces. Proceedngs of 2011 Internatonal Conference on Computatonal Scence and Its Applcatons, ICCSA 2011, pp. 258-262(2011) 7. Moschaks, I.A., Karatza, H.D.: Evaluaton of gang schedulng performance and cost n a cloud computng system. Journal of Supercomputng, 59, 975-992 (2012) 8. Song, X., Gao, L., Wang, J.: Job schedulng based on ant colony optmzaton n cloud computng. 2011 Internatonal Conference on Computer Scence and Servce System, CSSS 2011, pp.3309-3312 (2011) 9. Uskenbayeva, R.K., Kuandykov, A.A., Cho, Y.I., Kalpeyeva, Zh.B.: Tasks schedulng and resource allocaton n dstrbuted cloud envronments. Internatonal Conference on Control, Automaton and Systems, 16, 1373-1376 (2014) Copyrght 2016 SERSC 219