Fe24972976

251 views

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
251
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
2
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Fe24972976

  1. 1. Pankti Dharwa/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.972-976 Light-Weight Communication System in Grid Pankti Dharwa (Department of Computer Science,Sobhasaria Engg. College, RTU)ABSTRACT Grid computing is an interesting research area thatThe advantage of high-performance networks in integrates geographically-distributed computingconjunction with low-cost, powerful resources into a single powerful system. Manycomputational machines have made possible the applications can benefit from such integration.development of a new set of technologies termed Examples are collaborative applications, remote"computational grid". These technologies are visualization and the remote use of scientificmaking possible the creation of very large-scale instruments. Grid software supports suchdistributed computing systems by interconnecting applications by addressing issues like resourcegeographically distributed computational allocation, fault tolerance, security, andresources via very high-performance networks. heterogeneity. Parallel computing on geographicallyOne particularly difficult issue is that of utilizing distributed resources, often called distributed super-fully available bandwidth while computing, is one important class of grid computingbeing in some sense fair to competing traffic applications.flows. It has been widely demonstrated that TCP,the communication protocol of choice for most 2. RELATED WORKdistributed application, Based on the existing grid securityoften performs quite poorly in the emerging high- technologies, an authentication and access controlbandwidth high-delay network environments. framework at Virtual Organization (VO) level forThis has led to significant research on the group communication in grid environment. Thedevelopment to user-level applications that can implementation of prototype based on GridShib.circumvent some of the performance problems GridShib, developed to solve cross-domaininherent in TCP. The goal of this paper is to authentication and distributed authorization [1]. Thedevelop Light-Weight Communication System in system manages each part as a separate process (thethe Grid. The first initiative is to apply multi- single program, multipledata model) thatcasting features to this protocol & the algorithm communicates via messages [5].for the same is developed. Then apply the The Multi-Link Translator and Display Systemalgorithm on the Grid by using MPI (Message provides three basic capabilities: to receive andPassing Interface). Reduce the System call, which display tactical data link information, to translateis not necessary by the Windows at all the time between tactical data link message sets, and to routeand which consumes more time in background the data link information between various physicalprocesses. interfaces [6]. User management model developed for the lightweight multilevel grid system GrOSDKeywords – Alchemi Framework, Grid computing, (Grid-aware Open Service Directory). Main purposeLight Weight Communication, Message Passing in making the design was the simplicity andInterface. scalability [8].1. INTRODUCTION • Role-Based Access Control (RBAC), in group The main objective of this paper " Light- communication systems, there exist many kinds ofweight communication system in Grid" for building roles, such as group creator, group controller, manyan efficient, flexible, and scalable inter-grid kinds of application managers and applicationcommunication. To make a Communication Light- members. In which they used attribute-basedWeight in Grid, because of the limited network approach [1].bandwidth, high throughput, more reliability, byreducing the unnecessary System call and applying • Dynamic scheduling system communication,the Multi-Cast Protocol with MPI for the monitoring module monitor the running of sub-taskscommunication. Challenges in designing such in the simulation process of the operation, it can getinfrastructure arise due to the distributed nature of the CPU utilization rate, delay time and port traffic,the resources to be used, the distributed etc, also can display in real-time. Dynamiccommunities, the size of the data to be shared and scheduling module receives the monitoring resultsthe limited network bandwidth. information, and then under the dynamic algorithm migrate simulation task to achieve load balancing [3]. 3. EXISTING METHODOLOGIES 972 | P a g e
  2. 2. Pankti Dharwa/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.972-976 A grid application is defined simply as an grid head node(Manager), which contains the application that is to be executed on agrid and that services like message handler/dispatcher services, consists of a number of grid threads. Grid directory services, etc. Grid head node see the applications and grid threads are available resources, and send exposed to the grid application developer via the the query message or the application thread to other object-oriented Alchemi .NET API[4]. grid node. The grid executor or the execution node3.1 Basic Architecture of Communication System compute the application thread or the query message and give the output back to the grid head node(Manager), and that node collect the output of from all other grid execution node and combine it to gather as one and give it back to the application client or the user. Figure 1. Basic Architecture of Communication Figure 2. Normal Flow of Grids System 4. PROPOSED ALGORITHM AND METHODS 4.1 Optimization of Light-Weight The basic architecture of the communication system Communication is as described below. The user application is send to the Grid head node(Manager), which has container Optimize communication services is essentially any in type of semantics associated with communication which all layer for the communication system is beyond the simple, reliable unicast, transfer of data described as the Application mode llike ,thread, task, from point A to point B, or even the multicast of etc. Then the Allocation Manager check the data from one to many [7]. available resources and then the Message Handler/or Dispatcher, send or There are several ways to Optimize the Light- receive the message for the communication. There is Weight Communication.some ways are, security mechanism like, Authentication,Authorization is used for  Using multiple networks in parallel the communication according to the gird protocol.  Simplifying LAN-wide host naming And finally the Communication layer can communicate to the other Grid Execution  Simplifying communication protocol node(Executor) [9].  Avoiding temporary buffering of messages 3.2 Working Flow of Grids The normal flow of the Alchemi grid is as described  Pipelined communication path below. There is/are the user application or  Avoid system calls for communication application client sends the application thread or the query message to the 973 | P a g e
  3. 3. Pankti Dharwa/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.972-976  Lightweight system calls for 6. MPI Recv: reads the local receiving buffer. communication 7. MPI Bcast/Mcast: sends data to multiple  Providing very low-level mechanisms processes.4.2 Multi-Cast Protocol With MPI 4.3.1 Basic Concepts of Data Transfer • All MPI communication is based on communicator which contains a context and group. • Contexts can be defined as safe communication space for message passing and also it allows the different libraries to co- exist. • Group is set of processes. • Processes can be identify by unique rank in group. 4.3.2 Message passing is required for: • Data transfer plus synchronization • Requires cooperation of sender and receiver Figure 3. MPI using Multi-cast Protocol • Cooperation not always apparent in code The Multi-cast protocol for send/receive 5. IMPLEMENTATIONwhich combine with the MPI(Message Passing Alchemi is a .NET based grid computingInterface)algorithm, it combine with the windows framework that provides runtime machinery andgird by using API- GThread, programming environment required to construct theGApplication, GConnection to the Grid head node. desktop grid and developedGrid head have storageor the schedular which the grid application. It allows us flexible applicationschedule the application.and give it to the grid other composition by supporting the object-oriented gridnode to execute programming model. Cross platform support isand take the output or the values back from it. This provided by Web-services and flexible executionall process is done on the windows .Net framework. model supports dedicated execution by grid nodes [16].4.3 Message Passing Implementation Message Passing is a communication 5.1 Alchemi Frameworkparadigm to develop parallel and distributed The aim of Alchemi grid computing framework isapplications that require inter-process not to develop the grid software as easy as possiblecommunication. The Message Passing Interface but flexible, scalable, reliable, and extensible. The(MPI)computing models are specifications for key features of the Alchemi are,parallel that have numerous different • Internet based cluster computing for desktopimplementations [17]. Below are the main functions computer without a shared file system.i will implement for MPI. • Dedicated execution by cluster and individual1. MPI Init: message passing initialization method. nodes.2. MPI Finalize: message passing finalization • Object-oriented grid thread programming model.method. 5.2 Grid Thread Programming3. MPI Comm size: returns the total number of The two central classes in alchemi grid API areprocesses. GThread and GApplication that represent the grid4. MPI Comm rank: returns the process identifier. thread and grid application respectively. GThread is used for the code to be executed remotely and5. MPI Send: sends data to a remote node. GApplication is for locally executed code [16]. 974 | P a g e
  4. 4. Pankti Dharwa/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.972-9765.3 About the Bandwidth Monitor Bandwidth Monitor tracks traffic of allnetwork connections and displays real-timedownload and upload speeds in graphical andnumerical forms (refer to screen shotbelow). The software logs traffic of all networkconnections and provides daily, weekly and monthlytraffic reports. Bandwidth Monitor can also displaydownload and up-load speeds of multiple network connections on acomputer at a time. For example, you can monitorbandwidth of multiple network cards on yourcomputer at the sametime. Bandwidth Monitor also offers useful built-in Figure 5. Packets Sent/sec in Gridutilities: speeds stopwatch, transfer rates recorder,and bandwidth usage notification. 6.1.3 Packets Receive/sec in Grid6. EXPERIMENTALRESULTS AND ANALYSISThe result, have been obtain using the "PERFWIZ"simulator, and "Bandwidth Monitor" are shown in thetables. Table 1: Various Performance Parameters Total Avg Bytes Avg BytesExecutor execution sent(kbps) receive(kbps) Time(sec) 1 43.76 1.05 0.32 2 22.23 1.60 0.41 3 15.02 1.27 0.44 Figure 6. Packets receive/sec in Grid 4 15.01 1.72 0.54 7. RESULT ANALYSIS From the various kind of graphs, it shows the different parameter for measuring the6.1.1 Various Grid Nodes and Execution Time performance of the grid. In first Graph, shows that the execution time of the applica- tion or the job is decrease according to the increasing the no of grid execution node. In second and third graph, parameters like Packet sent/receive per second. From that, the difference of the execution time and other parameter increase/decrease accordingly with the no. of grid execution node(s). The Parameter’s value increasing with the no of grid nodes are, packe tsent/sec, the no of grid nodes are, application’s execution time, 8. CONCLUSION Light-Weight Communication helps in improving the performance parameters like Figure 4. Execution Time execution time, network bandwidth, memory usage, data sent/receive, etc. Since6.1.2 Packets Sent/Sec in Grid existing TCP/IP protocol when used in grid, does not provide, priority based routing, rate and burst control, multi-casting, etc. An attempt is made to overcome this and build the Light-Weight Communication Protocol. 975 | P a g e
  5. 5. Pankti Dharwa/ International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.972-976From the various result, it can be concluded that Georgiev, Kiril Boyanov, Maciejusing "Light-Weight Communication System in Malawski, Marian Bubak, Stavros Isaiadis,Grid", application’s execution time can be reduced Vladimir Getov, "USER PROFILINGup-to 15-25% of the normal execution time. Other FOR LIGHTWEIGHT GRIDS".identified parameters are volume of data transfer and [9] George Clapp, Joel W. Gannett, Ronaldthe network protocol used in Grid, specifically for Skoog,"Requirements and Design of athe multi-casting. The other parameters like network Dynamic Grid Networking Layer", 0-7803-bandwidth, 8430-W04F 2004 IEEE.network topology, the load on the network are very [10] Craig A. Lee, Eric Coe, B. Scott Michel,important for a grid to become as "Light-Weight James Stepanek,Ignacio Solisy, J. MattGrids". Clark, Brooks Davis,"Using Topology- Aware Communication Services in Grid9. ACKNOWLEDGEMENTS Environments" Proceedings of the 3rd I am Pankti Dharwa. I am pursuing MTech IEEE/ACM International Symposiumon(Computer Science) from Sobhasaria Engineering Cluster Computing and the GridCollege, Sikar. Rajasthan Technological University. (CCGRID.03),0-7695-1919-9/03 2003My research area is Grid Computing. I am very IEEE.thankful to faculty members and to my friends for [11] Jack Dongarra, Thomas Sterling, Horsttheir support. And at this moment, I would like to Simon, and Erich Strohmaier,"HIGH-express my appreciation to my family members for PERFORMANCE COMPUTING:their unlimited encouragement and support. CLUSTERS, CONSTELLATIONS, MPPS, AND FUTURE DIRECTIONS", 1521- 9615/05/ 2005 IEEE.REFERENCES [12] Ryan X. Wu, Andrew A. Chien,Matti A. Hiltunen, Richard D. Schlichting,[1] Deqing Zou, Laurence T. Yang, Weizhong Subhabrata Sen, "A High Performance Qiang, Xueguang Chen, Zongfen Han, "An Configurable Transport Protocol for Grid Authentication and Access Control Computing", 0-7803-9074-1/05/ 2005 Framework for Group Communication IEEE. Systems in Grid Environment", 21st [13] Marinho P. Barcellos,Maziar Nekovee, International Conference on Advanced Michael Daw,"High-Performance Reliable Networking and Applications(AINA’07) 0- Multicasting for Grid Applications", 7695-2846-5/07 2007 IEEE. Proceedings of the Fifth IEEE/ACM[2] I. Foster, C. Kesselman, L. Pearlman, International Workshop on Grid Computing etal.,"The Community Authorization (GRID04)1550-5510/04 IEEE. Service: Status and Future", In Proceedings [14] P. G. Viscarola, W. A. Mason,"Windows of Computing in High Energy Physics 03 NT Device Driver Development", OSR (CHEP ’03), 2003. Open Systems Resources, Inc. 1999, p10.[3] Fu Yanfang, Liu Bailin,"Research of [15] "A networking approach to grid Communication Technology for Simulation computing",Willey Edition,2004,Daniel Grid System", 978-1-4244-2800-7/09/ 2009 Minoli. IEEE. [16] Akshay Luther, Rajkumar Buyya, Rajiv[4] R. Buyya, M. Murshed, "GridSim:A Ranjan, Srikumar Venugopal,"Alchemi: A Toolkit for the Modeling and Simulation of .NET-based Grid Computing Framework Distributed Resource Management and and its Integration into Global Grids". Scheduling for Grid Computing", Concur- [17] Carlos Queiroz, Marco A. S. Netto, rency and Computation: Practice and Rajkumar Buyya,"Message Passing over Experience, vol.14,pp.1175-1220,November Windows based Desktop Grids". 2002.[5] Geoffrey Fox, "MESSAGE PASSING: FROM PARALLEL COMPUTING TO THE GRID" 1521-9615/02/ 2002 IEEE.[6] H. T. Ho, The MITRE Corporation, Bedford, MA, "Windows NT Multi- threaded Design in Multi-Link Translator and Display system", 0-7803-5749- 3/99/1999 IEEE.[7] "High Performance Cluster Computing, Architectures and Systems",Rajkumar Buyya.[8] Lazar Kirchev, Minko Blyantov, Vasil 976 | P a g e

×