A dynamic monitoring framework and methodologies adapted for Grid GARUDA

Abstract

This paper highlights grid operational processes, which are being adapted and practiced to monitor and manage the Indian national grid computing initiative GARUDA. GARUDA is an established grid, spread across more than 30 Indian cities and connected to more than 70 institutes. Various Virtual Organizations (VO) are formed for user communities to collaborate for the realization of common goals. However the resources in the grid may dynamically join or leave and may vary with respects to their administration membership. In this transient nature, the users must be able to keep track of resources and the grid operations are required to monitor and manage the shared resources (software, hardware, data and instruments) for efficient scheduling, data replication, accounting, performance analysis and optimization of resources. To achieve this, various open source and in-house developed tools are used. Monitoring broadly includes network health, cluster up time, middleware and related service status, host certification status, grid security, job statistics and analysis, resources status and so on. Problems are accordingly attended and based on their nature they are escalated to concerned teams, the National Knowledge Network (NKN) team and resource provider institute's Grid Administrators (GA). Various tools that are used for grid monitoring and operation are Paryavekshanam, Nagios, Link Monitoring Tool, Request Tracker etc. Most of the component failures are reported through alert emails to GA. Online reports are available for ease operation. Test scripts are installed at GARUDA Head Node (GHN) for automatic restarting of non functional components. All the results can be viewed online and data archiving is implemented for old data. We have found that proper monitoring methodology, highly transparent and interactive work culture and adhering to quality standards * gives good results in providing high grid availability and stability.