Information Service For Performance Prediction Of Grid Applications

Benefit from the conception of Electronic Grid, Grid computing wasgiven birth and developed rapidly in company with internet techniques.Various computing resources, storage, data and other specific resources,which are high-powered, geographically distributed and heterogeneous, areconnected in Grid computing schema via internet to achieve optimizedcomposition of resources dynamically and establish collaboratehigh-performance computing to solve huge application problemscooperatively, hence the name wide-area high-performance meta computingtechnology, a kind of distributed computing technology as well. A "VirtualSuper Computer" is composed of all kinds of resources with distinctionsamong hardwares, operation systems, organizations and regions masked byGrid infrastructure, shared and managed transparently and efficiently bydifferent users and organizations. Grid computing has powerful dataprocessing capability and is capable of making full use of idle computingresources through internet.Certain system structure is needed to build a Grid system. Grid systemstructure must identify the basic conponent parts of Grid, clearly explain therelationship among parts and their mode of integration, describe the function,purpose, feature and running mechanism of parts. Grid system structure is thecore technique of Grid, the skeleton and soul of Grid. A reasonable systemstructure can be used to design and build Grid, ensure the capability of Grid.Grid should accept all kinds of equipments, any one that follows Grid's rulescould join in Grid. Unified and standard interfaces are provided, which areindependent from the conditions of accessing equipments. User doesn't needspecific training or knowing of technical details to use Grid. Resources in theGrid can enter or exit without influence to the use of entire system. The most influencing Grid system structure is the WS-ResourceFramework. Grid needs to contain and use fuctions of web service. Thecapacity for user to accesss and operate the state-data is supplied by webservice. Standards of managing service's state-data need to be defined forapplication to discover, monitor and interact with stateful resources in astandard and operable way. The purpose of proposing WS-ResourceFramework is to standardize web service's capacity and introduce theconception "stateful" into web service. It means Grid group and web servicegroup develop on the same foundation.Scheduling is one of the core issues in Grid Computing. The approach ofeffectively discovering and querying resource information through differentadministrative domains is of great importance to the scheduling andexecution of Grid applications. Realization of task scheduling function baseson information service. Information service plays a decisively basic role inGrid computing environment, which supplies many basic informationoperations, including acquiring, publishing and managing. In the structure ofGrid information service, different information providers generateinformation of sorts according to resource properties.Monitoring and Discovery Service (MDS) is a component in GT4(Globus Toolkits 4.0). As an important part of information infrastructure, itmainly contains a series of network services based on monitoring anddiscovering mechanism. Monitoring and discovering mechanism cares aboutthe structrue of resources and services in the Grid as well as the acquiring,distributing, indexing, archiving and processing of state information, it is anactual and real-time reflection of Grid computing environment. Informationprovided by MDS is as follows: resources in the Grid environment, states ofthe Grid environment, execution of the Grid tasks.Resource monitoring is the base of resource prediction. Computing Gridneeds real-time, precise, dynamical performance information. Monitoringbegins with sensor, any program that generates performance monitoringevents with timestamp can be used as a sensor. Sensor collects variousdynamical information, such as CPU load, free memory space and networkstate. Resource prediction model is generated according to monitoringinformation, then time series of resource prediction is acquired to predict theusing of resource in a certain duration.Resource prediction (such as load information of applicable host) couldbe converted to runtime prediction of application. Refering to the dynamicalchange of resource, this method can adapt the dynamical and heterogeneousenvironment of Grid. Time series is provided in real time. Measurement andprediction of resource are separated from applications, and they are periodical.Appalications that share resources could share resource prediction.Application performance prediction solves problems with time-sensitivemodel and predicts application performance in the Grid environment usingdynamical and heterogeneous performance information. Prediction mustcombine with task's resource demand in order to figure out predictedexecution time as direction of scheduling. Corresponding methodes and toolsare needed in finding suitable prediction model and using this model toestablish a swift predicting system with low overhead;then the methodes areused in hostload prediction to predict task's execution time with hostloadprediction;finally, prediction information is produced and published to serveGrid users and scheduling programs.This paper is part of project "Real-time monitoring and predicting ofGrid applications' execution performance" which is funded by NationalNatural Science Foundation. A suit of methodes and prototype are establishedto: provide real-time execution prediction for Grid applications;directscheduling system;make the utilization of Grid resources more reasonable;guarantee the threshold value of task's finish time;improve the executionperformance of applications.Information service mechanism in Grid is explained, its structure isestablished using MDS. Prediction model of application performance isproposed consulting RPS prediction tool. An information service prototypesystem for performance prediction of Grid applications is designed thenestablished. The prototype system provides unified data type and friendlyuser interface, fills up the lack of application performance predictioninformation in Grid information service, and brings practical meaning toeffective scheduling, delivery and execution of Grid applications.The prototype system should be consummated and extended in thecoming future. In company with the thorough development of Gridtechniques, performance prediction information of Grid applications in otheraspects will be called for by Grid users and scheduling programs, suchinformation about memory, bandwith and hard disk volume that a task willprobably need. Applications will be sorted and modeled in different types,to enhance accuracy and adaptability of prediction information. Consideringincreasing number of nodes and sharing of multi-domain resources,centralized index service will probably be a main performance bottleneck inlarge-scale Grid environment;a distributed index directory will be needed toimprove the performance of information service.Grid is a new technology, related standard and criterion is beingestablished and improved. Functioning as the cornerstone of Gridinfrastructure, Grid information service deserves more applied research andpractice.