A Study On Distributed Cluster Performance Monitoring

Nowadays,distributed cluster systems become the primary solution ofthe high performance computing and enterprise data dealing with. They allhave the same advantages, for example: compatibility, scalability, highuniversality, low price and so on. Now, distributed cluster systems have beenused in weather forecasting, aerodynamic simulation test, biologyapplications such as molecular modeling. According to the recently released2007 TOP500 list, more than 80 percent of the systems are clusters. With thedistributed cluster architecture recognized by the people, the importance ofthe cluster performance monitoring is increasingly obvious. The research onthe cluster performance monitoring help users make better use of systemresourcesandoptimizeclusterexpansion.At present, the research and business development on the monitoring ofthe cluster systems have more and more achievements. They all have theirprominentfeature.In this paper, for the research and analysis on performance of thedistributed cluster system, we make use of the open source software"Groundwork Monitor Open Source"and the cooperation project of GridComputing and Network Security Laboratory of Jilin University andPlatform CO. Platform EGO is a Grid platform offering virtual andautomationfunctions,sharingalltypesofapplicationsandITresources.EGOintegrated with asynchronous notification mechanism for reporting the statusof the components, this requires the ability of real-time capturing of theclustersystems.In this paper, we introduce the structure of the"EGO"and focus on thenotification mechanism firstly. Then we designed a system for distributedcluster performance monitoring-EPM, and realized the function of EPMaccording to the overall structure of EPM. Finally, users can monitor theperformance of the distributed cluster systems in the help of this monitoring system. Absolutely, users also can monitor the performance of EGO likeother systems.This paper divided the monitoring system into three layers: monitorlayer, internal control layer, visual services layer.In the client, users can get the result of the monitoring and control thedetails of the monitoring from the web browser directly. Users can configurethe functions of the internal control layer through the web interface providedby monitor layer without knowing the internal logic of the internal controllayer.The internal control layer is the core of EPM. The responsibility of it isdata collection, data analysis, data saved, configuration of the monitoringsystem and so on. When it's collecting the data from monitor layer, itdiscoveries the exception from monitor layer, it will trigger the exceptionhandling. After it finishes the collection of the data, it saves the data intofixed format that convenient for express in the visual services layer.The monitor layer locates the bottom of the monitoring system. It'scomposed of external plug-ins. The function of the plug-ins is monitoring thedifferent resources. The format of the plug-in can be defined by users. Theplug-ins are compiled executables or scripts (Perl scripts, shell scripts, etc.).The mechanism of plug-in enriches the capabilities of the EPM andincreases the flexibility of the EPM.The design of the monitor of EGO performance based on the opensource software"Groundwork Monitor Open Source", and we developedsecondarily on the basis of it, as a result, shorten the development cycle.Groundwork Monitor provided the functions of user management, visualmonitor, personalized interface etc. These powerful functions will reduce thedifficulty of development. At other aspects, reference the notification of EGO,the author developed the monitor of asynchronous notification. This functionperfects the ability of the monitoring system.Generally speaking, the design idea and technology this paper presents mesh with the current direction of development and trends of the monitor ofdistributed cluster system, the practical use of the system also reflects thebasic structural, advantages, and there is a definite reference value in thedesign of distributed cluster system.