An Efficient Resource Prediction Model for Mobile Grid Environments Umar Farooq Punjab University College of Information Technology Lahore, Pakistan
[email protected]
Saeed Mahfooz Department of Computer Science University of Peshawar, Peshawar, Pakistan
[email protected]
Wajeeha Khalil Department of Computer Science University of Peshawar, Peshawar, Pakistan
[email protected]
Abstract- Grid Computing has emerged as an efficient problem solution paradigm since the last decade. Most of the research and implementation of grid computing environments collaborate and share heterogeneous resources in static manner. Scheduling; an activity of the scheduler is the most challenging and primitive issue that should be addressed effectively to achieve minimized resource utilization and maximum performance, reducing the average execution time. To achieve the above mentioned goal, many prediction models have been presented based on different parameters such as network bandwidth, computing capability and replica management. Recent years have seen a keen interest of research community and practitioners for the development and deployment of mobile grids. This paper aims to present an efficient mobility model to extend the conventional prediction models. The model visualizes best resources and hence improves scheduler performance in mobile grid environments. A sample implementation mechanism and two possible scenarios have been demonstrated. KEYWORDS: Grid, Mobile Grids, Prediction, Scheduling, scheduler
Decentralization,
I .INTRODUCTION
Grid Computing has emerged as a new computing paradigm in recent era for the solution of computational and data intensive problems. Grid computing leverages heterogeneous resources from different autonomous organizations based on some common usage and /or collaboration policies. Grid computing has become the first choice for problem solution in science, business, and commerce by adopting different features from existing paradigms such as cluster computing, utility computing, autonomous computing and peer-to peer computing. Grid computing provides a way to create dynamic virtual environment(s) called virtual organization(s) for the solution of problem(s). A virtual organization provides coordination for large scale resource sharing and problem solution based on common polices for the participants. Multiple administrative domains, autonomy, heterogeneity, scalability and dynamicity/adaptability are some important characteristics of a Grid.
ISBN: 1-9025-6013-9 © 2006 PGNet
Grid Computing has the capability to solve large scale problems by utilizing geographically distributed heterogeneous resources owned by different organizations. In grid computing, problem solution is initiated by user via Grid Resource Broker. Broker on the behalf of user in turn performs resource discovery, scheduling and processing of application jobs on distributed Grid resources [1]. The Producer registers the resource with GIS (Grid Information Service) by giving resource usage policies [2]. The Grid resource broker accesses the GIS to find suitable resources for the solution of user problem by fulfilling the user constraints (budget and deadline constraints) [2]. The development and deployment of a number of low level services and high level services/tools are required to build a grid. Resource management and scheduling are the most challengeable aspects of Grid Computing [3]. In addition, organizational policies, cost models, varying load and availability patterns introduce challenging issues as the producers (resource owners) and consumers (resource users) have different goals, objectives, strategies and requirements [4]. To process an application on grid, the problem is decomposed into segments. The segments are assigned to proper resources for processing. Proper resource selection from the available pool of resources is the most challenging task in resource management. Resource selection is a preliminary step to scheduling having high impact on scheduler performance. Different techniques have explored the best resource selection on the basis of historical information for static environments [3][5][6][7]. Recent years have seen a keen interest in the implementation of grid concepts in mobile environments. This dimension of research has given a new direction to grid computing from traditional static mode distributed computing to pervasive and utility computing based infrastructure. This infrastructure called mobile grid exploits the advanced capabilities of wireless networks and lightweight thin devices. Though grid computing integrates geographically
dispersed resources and users to create a dynamic virtual organization, most of the resources are static in nature. This construction uses Internet as base platform in general today. Mobile Grid inherits the concepts and capabilities of Grid with additional support for the features such as mobile users and mobile resources in seamless, transparent, secure and efficient way [8]. The processing environment is comprised of two basic entities; user (Initiator) and the resource participating in problem solution. The grid environment may be categorized as one of the possible categories; considering the status of user and resource. The status of user/resource may be static or mobile. The possible categories may be termed as Static-Static, Static-Mobile, Mobile-Static and Mobile-Mobile. Resource selection becomes more challenging when mobility is considered. So, it is required to take into account the mobility of users as well as resources in resource selection. Mobile grid integrates the grid and mobile networks, giving a wide range of applications. This integration exhibits one of the above mentioned structures. The applications of mobile grids include scientific, public services and commercial businesses [9][10][11]. Mobile grids integrate mobile devices such as laptops, PDAs (Personal Digital Assistants) and mobile phones. The mobility issue in grid environments has introduced new challenges to the research communities especially in the areas of scheduling, adaptation, security and mobility. Power consumption and size of devices are some other serious issues to be addressed. Secondary problems include small screen size and difficult input mechanisms. Peer-to-Peer computing provides many useful technologies and ideas for building scalable and reliable mobile grids [12]. Different prediction models based on different parameters are discussed in [13][14][15]. The most challenging problem in mobile environment is disconnection problem [11]. This paper concerns to devise an efficient mobility model to predict best resources. This paper does not target the above mentioned issues. Due to highly unpredicted environments, iterative based task assignment to resources is proposed. This is the matter of special concern when the environment is totally decentralized [16]. This paper presents an efficient mobility model for best resource prediction in mobile grid environments. The proposed model takes into account the mobility of user and/or resources with other parameters such as computational capability, storage and replica management. The rest of the paper is structured as follows. Section 1 discusses introduction and background information related to static and mobile grids. Proposed mobility model for mobile environments is presented in section 2. A simple implementation mechanism and its demonstration for two scenarios (Static-Mobile and Mobile-Mobile) are provided in section 3. Conclusion and future work conclude the paper presented in section 4.
II. PROPOSED MOBILITY MODEL This section presents the proposed an efficient mobility model for resource selection in mobile grid environments. In general, the behavior of the user and/or mobile device is highly unpredictable. For a moving resource, there are two possibilities. A resource is moving apart or towards the other resource. The extreme possibility exploits the mobility of both user and devices. The mobility of both participating resources with the other parameters for computation may be combined for best resource selection. The mobility factor is used to predict for the time a resource remains in the vicinity of the problem initiator. Task assignment is based on predicted time. To calculate different parameters, one may use relative signaling or location awareness of resources. The calculations are performed at the problem initiator. The model presented in this paper is explained in terms of location awareness. Each resource knows its location information at any instant of time. The model assumes average speed of a mobile terminal with which it changes its position. The mobility must assume the average speed for a mobile terminal with which it changes its position. The model is devised for two scenarios termed as STATICMOBILE and MOBILE-MOBILE. The first scenario considers static initiator and mobile devices such as cellular infrastructure as an example [17]. The second scenario believes that both initiator and participating devices are mobile. This way usually discusses a totally decentralized infrastructure having no static infrastructure [18]. Based on the application domain, conventional schedulers consider different parameters such as computation ability and/or storage capacity for job assignment. The work on the above mentioned issues with different parameters may be found in literature [19][20][21]. Resource capability is assumed sufficient to process job(s). This paper devises an efficient mobility model and uses it to predict for best resources. This model may be integrated with the scheduler of mobile environments to improve its performance. The model believes in the use of a few base or derived/calculated parameters for both user and mobile devices. The parameters model consumes and/or uses include User Range, Average Mobility and Time in Range. User Range is a given fact about the initiator coverage, in which user can communicate with mobile devices. Average Mobility, a derived parameter, represents the average mobility of a resource and/or user (based on user and resource mobility). Average mobility is calculated based on two recent communications between user/initiator and resource with respect to the user/initiator. Time in Range parameter shows the predicted time for resource availability
within the user’s range. The patterns/equations used to calculate Average Mobility and Time in Range are provided next. Average Mobility = abs (first history – second history) (1) The history is simply the distance between user and resource. This may be simply calculated by finding difference between the two recent interactions. The Time in Range, parametric value is calculated by Equation 2. The “Distance” is the net difference between the locations of user and resource (new location). Time in Range = (User Range – Distance) / Average Mobility (2) The number of jobs to be assigned to a resource in a single iteration is calculated by Equation 3. The parameter “No. of Jobs” is calculated by considering the values of “Time in Range” and “Job Completion Time” parameters. The value of “No. of Jobs” parameter is used for task assignment to a resource. No. of Jobs = Time in Range / Job Completion Time (3) The following rule is applied for the assignment of job (s) to a resource based on the value of “No. of Jobs” parameter. If Value (No. of Jobs) > 0 Then Assign job(s) based the value of No. of Jobs Else Reject assignment End If
parameter
The above model works well for a static user initiated scenario. The model needs extension for an environment having both mobile user and mobile devices. The model is extended by calculating “Average Mobility” based on two mobility factors. One mobility factor is considered for user and the other for the resource. The user average mobility is simply obtained by subtracting old location from the new location. The user maintains again two histories of the resource. The “Average Mobility” for MOBILE-MOBILE scenario is calculated by Equation 4, considering the histories of resource and the user average mobility. Average Mobility = abs (first history – second history – user average mobility) (4) The other parametric values are obtained through the above derived equations for “Time in Range” and “No. of Jobs”. The “Distance” in “Time in Range” calculation is now
the net difference between the new locations of user and resource. The model covers well for the extreme cases where both user and resource are moving apart; giving the most feasible predicted time. There are the cases in which the actual time a resource remains in range is more than predicted time. This situation is highly unpredictable so we leave this matter unattended in this paper.
III. IMPLEMENTATION MECHANISM AND ITS DEMONSTRATION This section presents simple setup to demonstrate the proposed mobility model and resource prediction. The implementation assumes that user and resource(s) know their location at any given time. This location awareness is easily realized through GPS (Global Positioning System) coordinates [22]. The implementation considers the coordinates of XY-Plane as GPS coordinates for simplicity. The environment is assumed to be fully decentralized for the demonstration. The user is considered static while the resources are mobile changing their position with respect to the initiator. The ranges of XY co-ordinates are considered between 0 and 500. The range of the user is assumed 200 meters. The user’s location is (200,200) and the first location of resource selected is (250,200) and second location is (300,200) simultaneously. To calculate histories, two distances from user to the resource by considering two locations of the resource are considered. First history yields 50 and the second yields 100. Use the Equation for “Average Mobility” with the determined histories to find out Average Mobility. The computed value of “Average Mobility” for the considered example gives 50. The absolute values are considered in all the cases. The positive mobility value shows that the resource is going far and negative mobility value means the resource is coming towards the user. The “Time in Range” value obtained by considering “Range in User” and “Average Mobility” is 2 units. The unit time is assumed sufficient to compute a subtask assigned to a resource. Based on this prediction model two of the sub tasks will be assigned to the resource, which will provide result before leaving the users range. To extend the implementation for an environment where the user and resources both are mobile, the implementation is extended as follows. Consider the same specification mentioned for STATICMOBILE scenario. Initial location of user is (200,200) and the resource location is (225,200). The first history is computed with the initial positions giving 25. The user and resource moved to new locations. The user’s new location is (175,200) and the resource new location is (250,200). The new distance from user to resource is now 75. The “Average Mobility” with respect to user and resource is now 75. The
time in range is now 1, which shows that only one job may be assigned to the resource. [6] IV. CONCLUSION AND FUTURE WORK This paper presented an efficient mobility model for resource prediction in mobile grid environments. Paper presented a simple implementation mechanism and demonstrated the mechanism for resource selection in two infrastructures; STATIC-MOBILE and MOBILE-MOBILE. First infrastructure is based on static user and mobile resource orientation. The other infrastructure considers mobile user and mobile resources. The infrastructure is demonstrated with simple experimental setup for resource selection and job assignment. The resources are predicted based on the time it remains in the range of the initiator. With the introduction of iteration based assignment, the extra time for re assignment of interleaved jobs is minimized as compare to other conventional techniques. Future work includes the application of proposed model for the remaining scenarios. The proposed model will be integrated with scheduler. This integration seems will improve scheduler performance. The proposed model is implemented with the concept of location awareness. Future work may include the implementation with relative signaling.
[7]
[8]
[9]
[10] [11]
ACKNOWLEDGMENT We are thankful to family members, friends, and colleagues for their support and help during our research work.
[12]
REFERENCES
[13]
[1]
[2]
[3]
[4]
[5]
Mark Baker, Rajkumar Buyya and Domineco Laforenza, “Grids and Grid technologies for wide area distributed computing”, Software - Practice and Experience, 2002. Karl Czajkowski, Steven Fitzgerald, Ian Foster and Carl Kesselman, “Grid Information Services for Distributed Resource Sharing”, IEEE Press, 2001. R. Buyya, D. Abramson and J. Giddy, “Nimrod-G: An Architecture for a Resource Management and Scheduling System in a Global Computational Grid”, May 2000, Beijing, China. Foster I, Kesselman C and Tuecke S., “The anatomy of the Grid: Enabling scalable virtual organizations”, International Journal of Supercomputer Applications, 2001. R. Buyya, D. Abramson and J. Giddy, “An Economy Driven Resource Management Architecture for Global Computational Power Grids”, Proceedings of the 2000 International Conference on Parallel and Distributed Processing
[14] [15] [16]
[17] [18] [19]
Techniques and Applications, June, 2000, Las Vegas, USA. K. Czajkowski, I. Foster, N. Karonis, C. Kesselman, S. Martin, W. Smith, and S. Tuecke, “A Resource Management Architecture for Meta computing Systems”, Springer Verlag Press, Germany, 1998. G. Heiser and F. Lam, “Resource Management in the Mungi Single-Address-Space Operating System”, Singapore, 1998. Antonios Litke, Dimitrios Skoutas and Theodora Varvarigou, “Mobile Grid Computing: Changes and Challenges of Resource Management in a Mobile Grid”, Environment Department of Electrical and Computer Engineering, National Technical University of Athens. Thomas Phan, Lloyd Huang and Chris Dulan, “Challenge: Integrating Mobile Wireless Devices into the Computational Grid”, Pervasive Computing Laboratory, The University of California, Los Angeles, Semiconductor Manufacturing, International Corp, Shanghai, China, and Vizional Technologies, Inc. Santa Monica. David C. Chu and Marty Humphrey, “Mobile OGSI.NET: Grid Computing and Mobile Devices”, DCS, University of Virginia, November, 2004. Sang-Min Park, Young-Bae Ko and Jai-Hoon Kim, “Disconnected Operation Service in Disconnected Operation Service in Mobile Grid Computing”, Ajou University, South Korea. Dejan S. Milojicic, Vana Kalogeraki, Ranajan Lukose and Kiran Nagarajal, “Peer to Peer Computing”, HP labs Technical Report, HPL-200257, 2002. Warren Smith, Valerie Taylor and Ian Foster, “Using Run Time Predictions to Estimate Queue Wait Time and Improve Scheduler Performance”, ANL and ECDC, North western University, Evanston. Allen Downey, “Predicting Queue Time on Space Sharing Parallel Systems”, 1997. Richard Gibbons, “A Historical Pro filer for Use by Parallel Schedulers”, 1997. Adriana and Ian Foster, “On Fully Decentralized Resource Discovery in Grid Environments”, DCS, The University of Chicago, USA. www.howstuffworks.com Andrew S. Tannenbaum, “Computer Networks”, 4th Edition, 2003. Hongtu Chen and Muthucumaru Maheswaran, “Adaptive Task scheduling Algorithms for MasterWorker Applications in Grid”, Advanced Networking Research Lab University of Manitoba, Winnipeg, Canada.
[20] [21]
Sang-Min Park and Jai-Hoon Kim, “Chameleon: A Resource Scheduler in a Data Grid Environment”, Sept, 2003. C. Waldspurger, T. Hogg, B. Huberman, J. Kephart and W. Stornetta, “Spawn: A Distributed
[22]
Computational Economy”, IEEE CS Press, USA, 1992. Ahmed El Rabbany, “Introduction to GPS: the Global Positioning System”, 2002.
