Performance Evaluation of Cloud Computing for Mobile Learning over Wireless Networks

Transcription

1 Performance Evaluation of Computing for Mobile Learning over Wireless Networks Khaing Sandar Htun Graduate School of Business Assumption University Bangkok, Thailand Abstract computing is the newest technology which is very convenient for users as it provides limitless storage via network based upon pay-per-use. Besides it also provides other services such as Infrastructure as a Service (IaaS), Software as a Service (SaaS) and Platform as a Service (PaaS). is very useful for institutions and organizations since it can be borderless, easily accessible from anywhere at anytime. Thus this technology is very beneficial to mobile learners as it helps eliminate problems of distance barrier and the access to the education in different geographical locations. Users can easily upload and download data in order to share to/from the centralized system via wireless network in real time. This paper will evaluate the performance of clouds for mobile learning based on cost-effective analysis. The analysis results that Google cloud outperforms comparable to other four clouds. Keywords Computing; Mobile Learning, Performance evaluation; Economical cloud I. INTRODUCTION computing is an evolving computing technology which centralizes data storage with remote servers hosted on internet. It offers unlimited data storage through network based on pay per use meaning cloud users only pay for what they use on the particular type of service availed. services the users to use software with no installation needed which cut expenses to buy software products, installations, operation, maintenance and upgrading of computers software. Moreover it is the easiest way to get data records or to browse information from anywhere all over the world with network connection []. Another research referring to different network such as CSMA Networks with Correlated Traffic can be found in [], []. The user only needs the network connections, or cellular signals. There are five vital features, four organization structure models and with three service models. First vital feature in the cloud model comprise of: Broad Network Access which is accessible through public network from anywhere with gadgets. Measured Services which control automatically and enhance sources and ability to accomplish the task at one extent of concept fitting to the different kinds of packages (e.g. data storage, internet speed, activation for user accounts and handling the data) []. On-demand Self-service - users can individually access with the availability of cloud technology without requiring the interaction of internal IT. Rapid Elasticity is one of the essential characteristics which allow the quick sharing of sources that is capable and flexible upon request. Resource Pooling which can be shared and maintained more efficiently due to centralized resources or data storage. There are four organization structure models of cloud computing. Public cloud is an open use which is accessible to the universal community. Independent organizations can manage and operate on their own. It can be found within the vicinity of the cloud providers. Private cloud is an exclusive use by a single organization and solely accessible to internal users within that particular organization. The organizations or a third party can manipulate in and out of the vicinity. Community cloud is used by organization with the same interest or purpose in the business. Hybrid cloud is a combination of more than one of the type of cloud that makes it unique. The three service cloud models are namely Infrastructure as a Service (IaaS), Software as a Service (SaaS) and Platform as a Service (PaaS). IaaS is a model that is providing the tool that sustains the process together with databases, hardware, storage, components of networking and cybernetic servers. Regarding ownership of tools and its storage, monitoring and maintenance, service providers are responsible. The user usually pays for what they avail. SaaS,

2 a step higher than PaaS, provides software and applications running on cloud infrastructure. These applications can be accessed from any location. PaaS provides a computing platform through the websites. It facilitates making applications via websites swiftly and easily at a reasonable cost for purchasing and maintaining the software or program and hardware [9]. Fig. shows the anatomy of the cloud computing. CLOUD limitless data storage Service Levels Software as a Service Platform as a Service Infrastructure as a Service Public Private Hybrid Deployment Models Community Fig.. Architecture Technical specifications of four different gadgets, namely iphone5, ipad mini, Samsung S, and Samsung Galaxy Note, are shown in Table and speed of five different clouds are provided in Table. There are five different clouds used in this research namely Amazon EC, Gigenet, HP, Gogrid and Google []. Fig.. Computing Anatomy (Adopted from Communications of the Association for Information Systems, Volume, Article, 7--0) [0] Throughput is the work done within a given time and measures the results based on the work carried out or processed. Performance is another way to gauge the computer productivity. It means the speed with which one or more set of batch programs operate with a particular workload or the number of times the user requests for responsiveness of the computer that is called as response time. Mobile learning is the system where everyone can learn through different gadgets over wireless technology [6], [7]. It employed for learning over cloud computing. Through mobile learning, learners can access information and learning resources from the centralized shared data storage at anytime and anywhere [], [8]. Mobile learning is the most convenient way of learning as well as the best solution for distance barrier to access the education in different geographical location. II. CLOUD ARCHITERCTURE Fig. shows the design of different gadgets operating mobile learning accessing data or information from cloud over wireless network. As different gadgets have their own data rate or speed, the promptness of accessing data or information from cloud could vary. However they all must have equal probability upon accessing into the cloud by computing, every device have the probability to access at 0.5. Probability of accessing data is equal to the number of ways it will occur over the total of outcome. TABLE. TECHNICAL SPECIFICATIONS OF EACH FOUR GADGETS iphone 5 ipad mini Samsung S III Samsung Galaxy Note II Data rate Mbps Mbps Mbps Mbps Internal Memory / Storage Data rate (Gbps) Cost ($) TABLE. 6//6GB storage; GB RAM 6//6GB storage; 5MB RAM 6//6G B storage; GB RAM 6//6G B storage; GB RAM SPPED AND MONTHLY COST OF DIFFERENT CLOUDS 9,06 0,9 0,0, III. PROPOSED QUEUING MODEL Fig. shows the proposed queuing model tested with the mobile learning object file size of 500 times larger file size than the tested file, 00kb [5]. The proposed queuing model has shown Poisson distribution which denoted λ whereas the data exponential service rate is µ. λ and µ represent iphone 5, λ and µ represent ipad mini, λ and µ represent Samsung S and λ and µ represent Samsung Galaxy Note.

3 λ λ λ λ μ μ μ μ Fig.. Proposed Queuing Model Regarding the input data arrival rate which is λ for each of the devices, it is very important to make proper assumption for λ since the volume of data arriving to the devices is unknown. However, to be practical, the data interarrival time should not be set to less than the server processing rate of device. For instance λ should be set to higher than µ. The capacity of iphone 5 to process the mobile learning object file is µ. The arrival rate denotes the data arriving per period of time whereas server processing rate is the average number of data to be processed in a certain period of time. Queue can be caused when the data arrival rate is very fast and the servers are slow. In this simulation, the Queue priority is set to first-come first-served (FCFS) basis. The input data of Poisson distribution rate and mean exponential distribution rate for the simulation are shown in Table. μ5 Simulations are run five times for different cloud parameters. Firstly, simulation is run with Amazon EC with the bandwidth of 9,06Gbps at $885. Secondly, simulation is run with Gigenet with the bandwidth of 0,9Gbps at $6.. Thirdly, simulation is run with HP with the bandwidth of 0,0Gbps at $00 and fourthly, simulation is run with Google with the bandwidth of,0gbps at $0. Lastly simulation is run with Gogrid with the bandwidth of 0.Gbps at $699. All clouds represent the cloud bandwidth as well as the cost which to be analyzed in terms of performance and cost-effectiveness. IV. DATA ANALYSIS AND RESULTS Arena Version.9 is used as simulation tool to carry out this research. Arena software is user friendly software as it can create and run the experiments on models of any system with required proper input data. In this research, simulation time was set to 500,000 seconds to run the proposed queuing model. Table 5 to Table 9 shows the result of simulation, Utilization (UTIL) which measure the ratio of time when the system is occupied. It must be in the range of 0- or 0-00% to avoid bottleneck. The result shows that the system is still abit far from the bottleneck. Mean Queue Length () which measures the number of waiting in a queue. Mean Waiting Time () which measures the average waiting time in each mean service time before moving to the server. Table shows the result of simulation on Throughput (THRU), which the work is done within an observed time, of each cloud. It also states the comparison of each cloud performance with cost and time. In this experiment, there is only slight difference in the throughput of each cloud upon the simulation. TABLE. INPUT DATA (POISSON DISTRUBUTION RATE AND MEAN EXPONENTIAL DISTRIBUTION RATE) FOR THE SIMULATION TABLE. PERFORMANCE ANALYSIS OF ALL 5 CLOUDS Poisson distribution λ = 6 s λ = 6 s λ = 8 s λ = 6 s Mean Exponential Distribution μ =.57 s µ =.57 s μ = 7. s μ =.57 s μ 5() = 6.6 μs μ 5() =.7 μs μ 5() =.6 μs μ 5() = 6 μs μ 5(5) =.5 s THRU,757,757,757,75,750 Performa nce/cost Performa nce/sec A. Performance Analysis of Amazon Table 5 shows the simulation result of Amazon. Throughput is,757 jobs per second while the utilization of the cloud is next to zero due to the high speed of the cloud comparable to the gadgets. The utilization of Samsung S is 89.58% which is near to bottlenecked situation because it has the slowest speed among other gadgets. This 5

4 situation may boggle down the system throughput indirectly. Other gadgets are far from bottlenecked situation. There is no mean queue length () at Amazon. Samsung S has mean queue length of jobs per second while the maximum queue length at the same reaches jobs seconds. The maximum will reflect a buffer size in order to handle this queue situation at least about 8.75 MB. Mean waiting time () of Amazon is zero which shows that there is no job waiting or no queuing at this due to its high speed. of Samsung S seems to have high figure compared to other gadgets because the massive volume of mobile learning data arrives at rate close to the gadget processing time and causes congested. TABLE 5. AMAZON CLOUD SIMULATION RESULT/OUTPUT B. Performance Analysis of Gigenet The simulation result of Gigenet is shown in Table 6. Gigenet has the same throughput as Amazon even the speed of two clouds is different. The result is not so much different from the first cloud but the utilization of Gigenet is lower than Amazon. This is because Gigenet can offer faster data rate than Amazon. TABLE 6. Amazon EC THRU ,757 UTIL (%) GIGENET CLOUD SIMULATION RESULT/OUTPUT Gigenet THRU ,757 UTIL (%) C. Performance Analysis of HP Table 7 shows the simulation result of HP and it has more or less the same result with the first two clouds. Again HP has the same throughput like the first two clouds. HP results a bit higher utilization compared to Gigenet because of the slower data rate per se. The rest of the output data for gadgets is identical to the first two clouds. TABLE 7. HP CLOUD SIMULATION RESULT/OUTPUT D. Performance Analysis of Google Google s simulation result is listed in Table 8. Google results are with lower throughput compared to the rest due to its lower data rate. The utilization of the Google is higher than others. Though other clouds have no mean queue length (), Google has. TABLE 8. GOOGLE CLOUD SIMULATION RESULT/OUTPUT HP THRU ,757 UTIL (%) Google THRU ,75 UTIL (%) E. Performance Analysis of Gogrid Table 9 demonstrates the simulation result of Gogrid which has the slowest speed of all clouds. The utilization of the cloud is very high unlike other clouds. It is close to bottlenecked situation because it is congested all the time. Besides the mean queue length () and mean waiting time () are also high and congested. 6

5 TABLE 9. GOGRID CLOUD SIMULATION RESULT/OUTPUT F. Performance Analysis of Samsung S Table 0 shows the maximum value of queue length and waiting time for all five clouds. for all clouds is quite high compare to other gadgets which reflect a buffer size in order to handle this queue situation at least about 8.75 MB. Maximum waiting time for all clouds seem really high which is almost 0 minutes due to the huge volume of mobile learning data arrives at rate close to the Samsung S gadget processing time and causes jammed. TABLE 0. Gogrid THRU ,750 UTIL (%) Maximum Queue Length Maximum Waiting Time PERFORMANCE ANALYSIS OF SAMSUNG S REFERENCES [] ABDULLAH ALSHWAIER, AHMED YOUSSEF AND AHMED EMAM, A new Trend for E-learning in KSA using educational clouds, Advanced Computing: An International Journal ( ACIJ ), Volume, No., January 0. [] PETER MELL. TIMOTHY GRACE. The NIST Definition of Computing, September 0. [] N.MALLIKHARJUNA RAO, C.SASIDHAR V., SATYENDRA KUMAR, Computing Through Mobile-Learning,, 0. [] Google Last Access, February 0. [5] CANDRA AHMADI. Performance Analysis of Mobile Learning in Wireless Network. [6] QIN SHUAI, ZHOU MING-QUAN. computing promotes the progress of M-learning. International Conference on Uncertainty Reasoning and Knowledge Engineering, Pages., 0. [7] ANWAR HOSSAIN MASUD, J. Y., XIAODI HUANG. Enhanced M-Learning with Computing:The Bangladesh Case Proceedings of the 0 5th International Conference on Computer Supported Cooperative Work in Design, Pages. 7, 0. ISSN //$ IEEE. [8] HONGYU ZHAO, YONGQIAN WANG, LIYOU YANG. Research on Distance Education Based on Computing. Pages. 6, 0. ISSN //$ IEEE. [9] ABDULAZIZ ALJABRE, Computing for Increased Business Value, International Journal of Business and Social Science Volume, No. ; January 0. [0] MARK A. ROSSO, BERNARD J. JANSEN. Communications of the Association for Information Systems The Berkeley Electronic Press (bepress), Volume 7, No. 6, Pages. 8, 00. [] ZHEFU SHI, CORY BEARD, KEN MITCHELL. Analytical Models for Understanding Misbehavior and MAC Friendliness in CSMA Networks, Performance Evaluation archive Volume 66 Issue 9-0, Pages 69-87, September 009. [] ZHEFU SHI, CORY BEARD, KEN MITCHELL. Competition, Cooperation, and Optimization in Multi-Hop CSMA Networks with Correlated Traffic, International Journal of Next-Generation Computing (IJNGC), Volume, No (0), November 0. V. CONCLUSION From the simulation result, there is not much difference in throughput. However it is found that Google is the best choice for the cost-effectiveness compared to other four clouds. According to the throughput obtained from the simulation, Google can perform 598 jobs for $ spent. In terms of performance over time all clouds have the same output regardless of the different speeds of cloud servers. Based on performance analysis, the fastest offered speed of cloud is Gigenet. Also based upon economical point of view analysis, the cheapest cloud is also Google. The capacity of the cloud for processing is very high compared to the communication devices therefore any impacts on the throughput will depend on the amount of traffic all gadgets generate. Future research may focus on the impact of a massive traffic generated devices over a low speed cloud. 7