, pp.427-432 http://dx.doi.org/10.14257/astl.2016.139.85 Research on Cloud Resource Scheduling Algorithm based on Ant-cycle Model Yang Zhaofeng, Fan Aiwan Computer School, Pingdingshan University, Pingdingshan, 467002 Henan province, China {Yang Zhaofeng} Abstract: we propose a cloud tas scheduling algorithm based on Ant-cycle in this paper, which changes the pheromone updating strategy of traditional ant colony optimization, and considers both the pheromone strength and path length of the individual when updating the pheromone concentration. As the experimental results show that, when accomplishing cloud tas scheduling in large scale, the Ant-cycle tas scheduling algorithm has faster speed, and more balanced scheduling result. Keywords: cloud computing, tas scheduling, ant colony optimization, pheromone 1 Introduction In cloud computing, tas scheduling is one of the ey issues. Many emerging disciplines apply their research findings into solving scheduling problem, such as genetic algorithm, neural networ, artificial intelligence and distributed study etc [1-3], all tae solving scheduling problem as their applied research field. In recent years, different tas scheduling algorithms emerged, especially for distributed dynamic load balancing, using of artificial intelligence model and distributed computing based on Agent all achieved good results [4-5]. The solving process of simulated annealing scheduling method is to see for a combined state, to mae the least target function value. Optimizing problems with solid annealing simulation is the global best algorithm in theory, which is because it can accept the bad energy value with a certain probability, so to jump out of the local minimum. However, because of the slow convergence rate, the simulated annealing scheduling method needs a longer computing time in the actual solving [6]. The tas scheduling algorithm based on the genetic algorithm is a searching algorithm based on nature genetics and gene evolution, which taes biological evolution as a prototype, has a good convergence, and is able to generate new result in the searching space with existing results. However, the genetic algorithm has the lower searching efficiency, easy to converge early, and the encoding method, group size, probability of genetic operators all need to further research [7]. The tas scheduling algorithm based on the ant colony optimization collaborates and communicates with each other through information, to form positive feedbac, so to gather the ants in multiple paths to the ISSN: 2287-1233 ASTL Copyright 2016 SERSC
shortest path. The major characteristic is that, seeing for the best path with positive feedbac and distributed collaboration. However, when in the larger tas scheduling scale, the ant colony optimization will appear the phenomenon of high pheromone concentration in the non-best path [8]. In this paper, we build the cloud tas scheduling algorithm based on the ant colony optimization, and solve the algorithm delaying problem made by non-ideal pheromone distribution with the model Ant-cycle. 2 Cloud Tas Scheduling based on Ant-cycle 2.1 Ant-cycle The ant colony optimization is a parallel optimization algorithm with strong robustness, and is applied into many fields. Put M ants into N random cloud nodes, in the process of the ant searching for the target node, the ant decides the shift direction according to the pheromone concentration in each path, and always moves towards the direction with higher concentration. In the initial phase, because there is little difference of the pheromone concentration in each path, the ants may choose the path randomly. Record the traveling path of the ant K with list (=1,2,, m),and adjust it dynamically according to the moving process of the ant, ) indicates the state transition probability of the ant choosing the city j as the target at time t,as is shown in formula (1): A 1 2 I 1 2 in t I in t ( ) ( ) na 0 indicates the next allowed node of the ant ; j A j A (t) indicates the pheromone concentration left in the path between node i and node j at time t; I indicates the initial information transferred from node i to node j, the information can be obtained from the problem itself; I 1 d is the prior value from node i to node j, d indicates the distance from node i to node j, when d is smaller, B I is bigger, and (t) is bigger. (t (1) 428 Copyright 2016 SERSC
1 is the information elicitation factor, reflecting the information amount accumulated in the path, and the guiding role in the moving process of other ants, indicates the relative importance of the path, the larger the value is, the more inclining the ant chooses the path which other ants have passed. is the expectation elicitation factor, reflecting the importance of the elicitation information when the ant chooses path, indicating the relative weight of the computing power forecast value. In actual computing process, if the remaining information amount on the path does not get handled, as the searching process of the ant carries on, more and more information amount on the path will cover the elicitation information, so when each ant finishes a path or all the n nodes get argotic, the pheromone needs adjusting with some strategies, and decreases gradually with time, we use following rules to adjust the pheromone on the path (i,j) at time (t+n): 2 ( t n) (1 ) (2) m 1 In which, indicates the pheromone exertion coefficient, then (1- ) indicates the pheromone remaining factor,in order to prevent infinite accumulation of the information, we limit the value range of is [0,1]. (t) indicates in the visiting process of the ant from time t to (t+n),the remaining information concentration in the path from i to j, also the pheromone increment on path (i,j) in this cycle, initial time ( 0) 0. In order to further increase the accuracy of the ant colony optimization in large scale searching, we build the pheromone updating strategy of Ant-cycle, as formula (4) shows. Q D 0 if ant others cross node ( i, j) In which, Q indicates pheromone strength, which impacts the convergence rate of the algorithm on some degree; D indicates total distance of the path ant finishes in this cycle. (3) (4) Copyright 2016 SERSC 429
2.2 Cloud Tas scheduling algorithm Next, apply the ant colony optimization in 2.1 into cloud computing. In the cloud computing architecture model of Map/Reduce, each unit in the cloud environment is made up of two parts, one is the separate main job scheduling node (Master Job Tracer), the other is an affiliated tas allocation node from each node colony in this unit (Slave Tas Tracer). Tae the slave node domain as an undirected graph G (V, E), in which V is the assemblage of all the slave nodes in the Area, E is the networing assemblage collecting each slave node, evenly divide the cloud computing networ into several sub-districts, and put equal ants in each district, ants in each group only search in their own district, to see for appropriate computing node, which is also to see for the best path in E, the metrics we need to consider include following parameters: Expected execution time: (a), indicating time consuming that computing t c d(a) resource in the end of path a handles such jobs; Networ delaying:,indicating the largest networ delaying generated by path a. Networ bandwidth:,indicating the largest networ bandwidth provided by path a. B(a) Combine the networ delaying of expected execution time, computing resource amount of i in the end of a in time t. Suppose the characteristic set of a virtual machine resource m VM i ( t, a) : indicates C c, c, c } (5) i { i1 i2 im In which, m=3, bandwidth. c im means a K diagonal matrix, indicating CPU, memory and 3 Experiment Result and Analysis In experiment, we set the tas number from 40 to 200, computing node number is 8. In order to show the difference, and we set larger in the gap of nodes Qos property, mainly including CPU, memory and networ bandwidth. Meanwhile, we choose traditional ant colony optimization and ACO based on Ant-cycle in this paper, execute by 10 times and tae average, the comparison of tas executive time consuming is shown in figure 1. 430 Copyright 2016 SERSC
6.0 5.0 4.0 ACO Method Proposed Method Time(s) 3.0 2.0 1.0 0 50 100 150 200 Tas numbers Fig.1. Comparison of tas executive finishing time of two algorithms It can be seen from the two curves in figure 1, in traditional ACO and Ant-cycle algorithm in this paper, there is little difference in the time consuming when the tas scheduling scale is small. But as tas scheduling scale becomes larger, executive time of traditional ACO increases continuously, while the executive time of Ant-cycle algorithm increases little. It proves the speed advantage of the algorithm in this paper on large scale tas scheduling. 4 Conclusions We build a tas scheduling algorithm of Ant-cycle. By changing the pheromone updating strategy of traditional ACO, As the experimental results show that, compared with traditional ACO, the Ant-cycle tas scheduling algorithm is faster in large scale cloud tas scheduling, and the tas loading on each node is more balancing after scheduling. References 1. Yeo, H.-s., Phang, X., Lee, H., Lim, H.: Leveraging client-side storage techniques for enhanced use of multiple consumer cloud storage services on resource constrained mobile devices [J]. Journal of Networ and Computer Applications, 43, 142-156(2014) 2. Gani, A., Moatder, N.G.: A review on interworing and mobility techniques for seamless connectivity in mobile cloud computing [J]. Journal of Networ and Computer Applications, 43, 84-102 (2014) Copyright 2016 SERSC 431
3. Bayyapu, K.R., Fischer, P.: Load scheduling in a cloud based massive video-storage environment[c]. Proceedings-16th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, 349-356(2014) 4. Amoretti, M., Grazioli, A., Zanichelli, F. : Towards a formal approach to mobile cloud computing[c]. Euromicro International Conference on Parallel, Distributed, and Networ- Based Processing, 743-750(2014) 5. Bilgaiyan, S., Sagnia, S., Das, M.: A multi-objective cat swarm optimization algorithm for worflow scheduling in cloud computing environment[j]. Advances in Intelligent Systems and Computing, 1,73-84(2015) 6. Rahimi, M.R., Ren, J.: Mobile cloud computing: a survey, state of art and future directions[j]. Mobile Networs and Applications, 19(2),133-143(2014) 7. Gsior, J., Seredysi, F.: A decentralized multi-agent approach to job scheduling in cloud environment [J]. Advances in Intelligent Systems and Computing, 322, 403-414(2015) 8. Yoo, J., Kim, J.: The advanced Korea-computer access assessment system(k-caas) on smart mobile cloud environment[j]. Multimedia Tools and Applications, 31(5),128-136(2014) 432 Copyright 2016 SERSC