Performance Monitoring And Analysis Of Grid Applications

Grid is a new concept, it is a representative of advanced technology andbase facility, and it is high technology developing subsequently toInternet .With rapid developing of Grid, it provide a platform to shareresource. As the conventional distributing compute, resource manage andtask schedule is key in Grid , In a grid environment, the state of resourcesvaries dynamically, real time, accurate, dynamic performance information isneeded for grids. It is important for load management systems to choosesuitable resources from the resource pool that varies dynamically. But GridPerformance Monitoring affect resources manage and task schedule. GridPerformance Monitoring in Grids is the base for the scheduling system, itmakes utilization of the resource more suitable and improves the executingtime of applications.Grid Performance Monitoring plays a decisive role in Grid computingenvironment, which is a fundamental basis. The functions and characteristicsof the system should contain: It run in every node in Grid, and collect systeminformation (EMS memory, system load, available CPU), task information(the number of task, running circs, task state, etc), safety information etc;Itanalysis, deal with and store these information;And these information will beshow in some mode.This paper is part of project "Realtime monitoring and predictingexecution performance of Grid application" which is funded by NationalNatural Science Foundation. The project intends to establish a suit ofmethodes and prototype in order to: provide realtime execution predictionof Grid application as the direction of scheduling system;make the utilizationof Grid resources more reasonable;guarantee the threshold value of task'sfinish time;improve the execution performance of application.Application performance prediction solve problems with time-predictionand to predict the application performance in the grid environment usingdynamical and heterogeneous performance information. Prediction must becombined with the task's resource demand in order to figure out predictedexecution time as the direction of scheduling. Respective methods and toolsare needed in realizing such prediction to find suitable prediction model anduse this model to establish a swift predicting system with low overhead;thenthe methodes are used in hostload prediction to provide a way that task'sexecution time is predicted using such hostload prediction;finally predictioninformation are produced and published to serve grid users and schedulingprograms.Application performance monitoring and analysis in computational gridenvironment is one of the Grid performance monitoring, and is basis of theApplication performance prediction, It provide performance data, we will analysisdata and confirm performance bottleneck and improve performance 。We have done research on mechanisms about Grid service and gridarchitecture and grid performance monitoring, proposed an applicationperformance monitoring and analysis model based on analysis and practiceNetlogger and nlv tool. We have set up a prototype system of applicationperformance monitoring and analysis service, it provides unified data typeand realtime monitoring , makes a great improvement to Grid service onlacking of application performance monitoring, and brings practical meaningto effective scheduling, delivery and execution of applications in suchenvironment.The Netlogger application was developed to help provide a method tomonitor distributed network applications. In this article ,Netlogger pick upapplication performance data as sensor via Netlogger API inserting thingthen it record some event log to local log file ,and analysis logdata with visualization tool NLV, it will be show in point, load line ,life line.In Grid, main server draw log data in Grid registration user with netlogdfunction and nlforward function. Nlv show information, this implementsimple thing check and the compare of thing time in application.Authentication is a kind of security mechanism of grid. We utilize CA ofGlobus, and use a machine in the LAN as authentication server which canreceive the request from customs. Customers send their requests to rootserver which can not identify root user, so it must be request by general user.It realizes that every user has unique ID in the LAN. Then the server receivesand verifies the request, and sign CA which encapsulates period of validityand public key to users. If you want to submit jobs to broker through proxy,CA is an essential to create proxy which includes authorization informationand security policy. Before submitting a job, users must create a proxy whichis identified by resource broker,and running every monitoring task is onesubmitting job.Now,application performance monitoring have been complete tosubmitting job. In Grid .with increasing of the data number, the answer timewill be slow. It will affect Grid server performance. So the future work willcombine the monitoring data with the Grid information server, we willdemand and analysis the monitoring data to index server in convenience, itwill reduce answer time.The rapidly developing of Grid technology provide more and moresatisfactory service .It is our working direction to further study and practice..