Design and Implementation of Resource Monitor Module in Meteorological Computational Grid Platform

Computing resources, aggregated by meteorological computational Grid, are composed of high performance computers and storage resources.These resources are installed in different areas with different system structures, running conditions and workloads. In order to monitor the status of resources in meteorological computational Grid and to provide users and administrators with reference information, resource monitor module is designed and implemented as part of meteorological computational Grid platform software system.The resource monitor module involves 3 layers : remote meteorological Grid nodes, resource state information acquisition, and web representation.Resource state describes the system information of high performance computers in a meteorological Grid node, comprising 3 major parts, overall information, nodes information and jobs information. The layer of resource state information acquisition is made up of poller, collector, feeder and related configuration files.Correspondingly, the acquisition process of resource state information in the resource monitor module can be divided into 3 parts, polling, collecting and feeding. Web representation layer is on the top and provides users with resource state information through commonly used internet browsers.The resource monitor module is developed based on Grid management software UNICORE and client software ARCON Client, and implemented with Java and XML technology. ARCON Client Toolkit is used to implement node accessing function in the resource monitor module. The poller submits querying jobs for status information to computers in Grid nodes automatically and termly, and pushes it into the log queue when a job is submitted.The collector reads the queue and retrieves results of query. The feeder parses the results and writes a specially formatted XML file. The code of querying and retrieving is asynchronous so as to avoid waiting in querying. As a result, the monitor program runs stably and robustly. Major packages of resource state information acquisition layer are base driver, job scheduling, log queue query, resource state parsing, and configuration setting etc. The web representation reads the XML file containing the resource state query results, and implements resource state displaying via Flex and J2EE technologies.At present, 10 high performance computers have been brought into centralized monitoring in National Meteorological Information Center, Beijing, Chengdu, Guangzhou, Shenyang Regional Centers as well as Anhui Province. Resource monitor module is one of the key parts of meteorological Grid platform software system and providing real time services.In the future, with the further construction of meteorological computational Grid, the resource monitor module will see further application and put major computing resources in meteorological department into supervision.