2012an20
1 like251 views
This document describes a report submitted for a seminar assignment on dynamic load balancing in grid computing using a multi-agent system and tree structure. It was submitted by Vishnu Kumar Prajapati in 2013 for his M.Tech. in Advanced Networking at ABV Indian Institute of Information Technology and Management in Gwalior, India. The report discusses the historical background and motivation for grid computing and load balancing. It then reviews literature on dynamic load balancing policies, multi-agent systems, grid computing service architectures and objectives. The methodology section proposes reducing communication overhead compared to previous pool-based approaches to improve grid system performance.
2012an20
- 1. Dynamic Load Balancing in Grid Computing with Multi -Agent System Integration by Using Tree Structure A Report submitted for seminar assignment M.Tech. in ADVANCED NETWORK by VISHNU KUMAR PRAJAPATI - (2012AN20) ABV INDIAN INSTITUTE OF INFORMATION TECHNOLOGY AND MANAGEMENT GWALIOR-474 010 2013
- 2. Contents 1 INTRODUCTION 1.1 Historical Background of the Grid Computing 1.2 Load Balancing in Grid Environment . . . . . 1.3 Goal of Load Balancing . . . . . . . . . . . . . 1.4 Type of load balancing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 3 3 4 2 MOTIVATION 5 3 LITERATURE REVIEW 3.1 Dynamic Load Balancing Policies . . 3.2 Multi-Agent System . . . . . . . . . 3.3 Grid Computing Service Architecture 3.4 OBJECTIVES . . . . . . . . . . . . 6 6 7 7 8 4 METHODOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 5 POSSIBLE SOLUTIONS 12 6 CONCLUSION 13 1
- 3. List of Figures 1 2 3 4 5 Grid Structure Environment . . . . . . . . . . . . . . . . . . . . . . . Grid Computing Service Layered Architecture . . . . . . . . . . . . . Comparison between Existing policy and proposed policy . . . . . . . Grid Structure Environment . . . . . . . . . . . . . . . . . . . . . . . Combining Architecture for Grid Load Balancing with service model . 2 . . . . . . . . . . 6 8 9 10 12
- 4. 1 INTRODUCTION The Grids can be defined as services that shares computer power and data storage capacity over the Internet and Intranet. It is not just simple communication between computers but it aims finally to turn the global network of computer into a huge computational resource. It can coordinate those resources which are not subject to centralized control. The grid is to use standard, open, general-purpose protocols and interfaces. The grid is to deliver nontrivial Quality of Service. A computational grid environment behaves like a virtual organization consisting of distributed resources. A Virtual Organization is a set of individuals and institutions defined by a definite set of sharing rules like what is shared, who is allowed to share, and the conditions under which the sharing takes place. A number of Virtual Organizations exist such as the application service providers, storage service providers, but they do not completely satisfy the requirements of the grid. Grid computing focuses on dynamic and cross-Organizational sharing, it enhances the existing distributed computing technologies. 1.1 Historical Background of the Grid Computing Technology Networked operating systems Distributed operating systems Heterogeneous computing Parallel and distributed computing Grid computing year 1979-81 1988-91 1993-93 1995-96 1998 Table 1: history of grid computing 1.2 Load Balancing in Grid Environment n a Grid environment , there are several Load Balancing techniques such as Randomized load balancing, round robin load balancing, dynamic load balancing, hybrid load balancing, agent based load balancing and multi-agent load balancing . Round robin and randomized load balancing are simple and easy to implement. Dynamic, hybrid, agent base and multi-agent based load balancing are going to improvement or new ones introduced in grid load balancing solution. 1.3 Goal of Load Balancing The Goal of load balancing is that the workload is fairly distributed among the nodes and that none of the nodes are overloaded or under loaded. So that the computing power fully utilize from the multiple hosts without disturbing the user 3
- 5. 1.4 Type of load balancing there are two types of load balancing strategies called static load balancing and dynamic load balancing - Static load balancing makes the balancing decision at compile time and it will remain constant. In dynamic load balancing makes more informative decisions in sharing the system load based on runtime. the dynamic load balancing provide better performance compare to static load balancing. Dynamic load balancing classified into centralized approach and decentralized approach. In Centralized approach is managed by central controller that has a global view of load information in the system which is used to decide how to allocate jobs to each other. Another one decentralized approach all joints nodes are involved in making the load balancing decision. In the grid computing is the method based on collecting the power of many computers, in order to solve the large-scale problems; On the other hand, it offers to share hardware and software grid resources. So that maximizes the overall grid performance. Tree base infrastructure is focusing on the load balancing algorithm for the grid computing services (GCS). The main goal of the design to submit their computing task simply by having access to our grid computing service web site(GCSWS)and another objective of GCS to access the powerful computers or expensive software with very low cost to the our grid users. 4
- 6. 2 MOTIVATION The distributed computing technology are use to share the resources between the institutional, by using grid computing it will give more better performance them existing distributed computing technology. Currently, Grid computing technology can be used to connect heterogeneous computing resources to each other in a way that user can regard all of this structure as a single machine on which we can run very highly complex and massive application programs that require a high processing power and huge volume of input data. The grid computing systems have improved the throughput and increase the performance to the individual nodes and whole grid system by using the load balancing. So the load balancing in the grid system has a big role for utilization of the resource and reduced the response time. 5
- 7. 3 LITERATURE REVIEW Decentralize load balancing approach are based on redistribution of tasks among the available processors. The processors which is overloaded are transfer the tasks to the under load processors, by using High Level Architecture (HLA) environment. This process work at the run time, so generally there is none of the nodes are heavily loaded. 3.1 Dynamic Load Balancing Policies here are four type of load balancing .which consists of Transfer policy, Selection policy, Location policy and Information policy. Figure 1: Grid Structure Environment • Transfer Policy: Transfer Policy should be transfer the load or not and it is based on various criteria such as workload value and computing Power. If the load balancing is needed it will sent to the selection policy. if not, the job will process locally. 6
- 8. • Selection Policy: The tasks define that it should be Transference or migrated from overloaded resources (source) to most idle resources (receiver). The decisions made by selection policy are then directed to the location policy for further process. • Location Policy: Location Policy are Uses the results of the Selection policy to find a suitable partner for a Sender or receiver. • Information Policy: In the information policy, the worked as what workload information to be collected, when it is to be collected and from where it is collected. 3.2 Multi-Agent System An Agent is a computer system that has a capability of taking independent action on behalf of its user or owner. The Multi-agent system hold several characteristics such as autonomy, local views, cooperation, social ability, reactivity, proactive, goal oriented and decentralized. Multi-agent system consists of communication layer, coordination layer and local management layer. The communication layer provides an agent with interfaces to heterogeneous networks and operating systems. It will receive the request and then explain and submit to the coordination layer to decide the suitable action according to its own knowledge. The local management layer performs functions of an agent for local grid load balancing. A Multi-agent system is composed of multiple intelligent agents that have the ability to interact or communicate, collaborate and negotiate among them. 3.3 Grid Computing Service Architecture Grid computing service (GCS) is allows to submit their computing tasks along with required hardware or software resources. It allocates tasks to the available resources and then executes the tasks. After execution, grid computing service will reply to the user and send back the results. As the following figure, GCS have four layers Web Service Task Submission layer, Grid Resource Monitoring layer, Task Allocation and Load balancing layer and Grid Task Execution layer. In the Web Service Task submission layer, work with user tasks submission and their requirements (resources and quality of service information). In the Grid Resource Monitoring layer, need to monitor those resources which are underutilized or overloaded. Each grid entry point is called a Grid Agent Manager (GAM).in the load balancing layer, there are two level of load balancing which are worker layer and GAM level load balancing. And the last Layer is Grid Task Execution layer, it is mainly responsible to perform tasks executions and also update the status of the hardware and software resources at a given computing unit. 7
- 9. Figure 2: Grid Computing Service Layered Architecture 3.4 OBJECTIVES To reduced the communication between worker nodes and Leader nodes and also between the Leader nodes. So that reduced the overhead compare to pool based approach and do the efficiently load balancing process. The main objective is to increase the performance of the Grid system, maximize the overall system throughput, minimize the response time and allow the good grid resources utilization. 8
- 10. 4 METHODOLOGY Figure 3: Comparison between Existing policy and proposed policy The information policy has making a decision and lot of contributions. We can say that information policy has a big implication on performance in grid computing through accurate, efficient and suitable for taking a decision. The transfer policy and selection policy are combined which is known as migration policy. By combining these policies, reduced the internal communication between policies in the agent as showing the above figure. Agent have a multifunction capabilities due to the role of embedded them. It will be two statuses which are leader of the computing element and worker of the computing element. The agent is automatic determined statuses or role themselves. If the agent is leader, it wills auto-notify the workload system manager. It also has the capability to communicate among the agent and exchange the information. The main work of the migration policy is receiving the data or if already holding the data, it will analyze the load and decide where process is locally or remotely. The decision made by the migration policy will submit to the location policy for further processing. Here the load balancing function work globally or locally. The load balancing decision making by the workload system manager which sits at top of the grid as described in the following figure. The workload system manager makes the decision based on computing element power or index and also to allocate the correct load value the correct computing elements which are the leaders in the local grid. Then, the computing element leader will decide how 9
- 11. Figure 4: Grid Structure Environment to distribute the load according to the worker node available computing power. Each worker node has the capability to auto-notify to the leader on itself computing power information, so that reduce the communication overhead compare to polling method. Load Balancing Algorithms: • APC=PC*L GPC =is the maximum processing capacity (tasks/seconds) at grid threshold utilization. So, AVGPC=GPC-APC All the above parameter are dynamic nature. APC=Actual Processing Capacity, GPC-Grid Processing Capacity, AVGPC=Available Grid Processing capacity. • Worker Level Load Balancing: If N is the number of received tasks at a given GAM(General Agent Manager), we define the following parameters- Where TPC= Total Processing Capacity, TAPC =Total Actual Processing Capacity. • GAM level Load Balancing: The GAM have to managed by tree structure in grid, the tree structure is selected to ensure the scalability (add/remove GAMS) and 10
- 12. minimize the communication between the GAMS. it also ensure that only one load balancing operation work at a time, so that ignore the inconsistency or wrong load balancing operations. by circulating the token message between GAM in the whole tree for exchange the information. The token message contains the global view of the grid system. it contain the following information about each GAM. Manager ID ,Total Available Processing Capacity(TAPC) of the GAM, status, Neutral(N) ,Receiver(R) and Sender(S) . 11
- 13. 5 POSSIBLE SOLUTIONS Figure 5: Combining Architecture for Grid Load Balancing with service model The user can be submitting a task to the grid web service. The user may choose the deferent web browser though web server to submit the task and also responsible to forward the request to the Grid Resource Monitoring Layer. The Grid Resource Monitoring Layer do the monitoring in heterogeneous resources like different Processing Power, different Internet speed, and the systems are in distributed manner, after that this layer work the central Workload system Manager for doing the accurate load balancing and task allocation (as by Bakri Yahaya ijincaa,2011) and forward the process to the next layer for execution of task. Workload system Manager have a Multi- Agent, An agent will determined what they are and automatically turn themselves into the determined status or role. If the agent is a leader, it will auto-notify the workload system manager. The agent itself has the capabilities to communicate among the agent and performs the information exchange. The global load balancing decision will be made by Workload system manager and the local load balancing will be made by leaders. The Grid Task Execution layer work as existing architecture (as by Abderezak Touzene IJCSI 2011). 12
- 14. 6 CONCLUSION In the Tree base architecture for grid computing services and Multi-agent system will reduced the internal communication. We also apply the load balancing policy to reduce the communication between policies. The workload system maintains the consistency and removed the wrong load balancing. By combining the policy method and grid computing Service Architecture we can achieve the maximum throughput, minimize the overall tasks response time and finding fault. In future we can apply the fault tolerance in the grid load balancing strategy. 13
- 15. REFERENCE 1. Sakadasariya Achyut R,”Survey of Resource and Job Management for Load Balancing In Grid Computing”. of the IJISME ISSN: 2319-6386 vOLUME-1 iSSUE-3, 2013. 2. S. Gokuldev, Shahana Moideen,” Global Load Balancing and Fault Tolerant Scheduling in Computational Grid”. of the International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 11, May 2013. 3. Preeti Gulia,Deepika Nee Miku, Analysis and Review of Load Balancing in Grid Computing using Artificial Bee Colony, in Proc. of IInternational Journal of Computer Applications (0975 8887) Volume 71 No.20, June 2013 4. Leyli Mohammad Khanli and Behnaz Didevar, A New Hybrid Load Balancing Algorithm in Grid Computing Systems, IJCSET, E-ISSN: 2044 - 6004 ., 2011 . 5. Bakri Yahaya, Rohaya Latip, Mohamed Othman, and Azizol Abdullah, Dynamic Load Balancing Policy with Communication and Computation Elements in Grid Computing with Multi-Agent System Integration, International Journal on New Computer Architectures and Their Applications (IJNCAA) 1(3): 757-765 The Society of Digital Information and Wireless Communications, 2011. 6. Abderezak Touzene, Sultan Al-Yahai, Hussien AlMuqbali, Abdelmadjid Bouabdallah, Yacine Challal, Performance Evaluation of Load Balancing in Hierarchical Architecture for Grid Computing Service Middleware,IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 2, March 2011. 14