Pfad/Path D-Grid gGmbH / Projects / Gap Projects / D-MON / 
Abb 1.: Implementierte D-MON Architektur

The vision of D-MON is to realise a grid-wide monitoring architecture above multiple, heterogeneous underlying systems. The difficulty in carrying this out lies in the fact that the D-Grid consists of many autonomous resource providers and virtual organisations. All middleware installations used in the D-Grid provide independent monitoring and information systems that can be differentiated by their functions, information models, data schemes, the implemented standards and their architecture. These systems merely coexist at present and are not interoperable.

The D-Mon project has made it its mission to lay the foundations for a D-Grid-wide monitoring system that works with the various standards as well as external tools. This is essential for monitoring the status of grid resources, services and jobs that are running in the various middleware environments. This is as relevant for users of the grids and administrators in the individual computing centres as it is for a Grid Operations Centre (GOC).

One aspect of D-MON is the horizontal integration of various grid monitoring systems. Problems that have to be dealt with here are incompatible interfaces, non-uniform software architectures, information models and schemata that are used by the established monitoring systems (Globus MDS4, gLite BDII and UNICORE CIS). In order to realise a usable cross-middleware monitoring platform, D-MON is relying on a central database for storing monitoring data. In principle, this monitoring data is extracted from each available information system by adapters and transformers and then added to the database. Access to the compiled monitoring data is to occur via a grid-compliant interface for databases (OGSA-DAI).
A close link between the D-Grid-wide monitoring system and the local management systems used by the resource providers should simplify the day to day work of the system administrators. For this reason, an additional adapter will be implemented. This will integrate the external administrative system and monitoring tools with the grid monitoring. This will be implemented using the example of Nagios, a tool for monitoring clusters.

D-MON will provide the communities and virtual organisations (VOs) with a VO-specific view of their resources and services only. Because of the large number of communities and resources in the D-Grid, this is absolutely necessary in order to maintain a clear overall picture. Otherwise, Grid users would be faced with mountains of irrelevant information since they do not possess the access rights for the respective resources or services.

The vertical integration in D-MON involves integrating VO-specific information in the monitoring architecture as well as providing this information. This vertical integration also encompasses the development of GridSphere portlets for displaying collected monitoring data in a user portal which, in addition to VO-specific views, will contain various roles for administrators and a GOC.

As shown in the figure, the resulting system provides dynamic, VO-based and user specific views in the services and resources provided. This can be achieved by linking the VO management information with the monitoring information provided by the information systems (CIS, MDS 4 and BDII) of the middleware installations. The combination of this data in a join, followed by a filter step, makes it relatively simple to create views in virtual organisations. Adapters that connect the various information systems and architectures facilitate grid-wide, cross-middleware use. These components (a database component on one hand and an adapter for connecting monitoring and VO management systems on the other) will be developed in the D-Mon project in order to realise the D-MON vision.

Another planned function that unfortunately cannot be realised in the D-Grid yet due to the lack of a corresponding authentication and authorisation infrastructure (AAI), is the integration of SAML-VOMS authentication/authorisation in the user portal and the OGSA-DAI interface. This negates the need for users to log-on manually or to select their own VO to determine the role in which they access the portal. The corresponding view can be generated automatically.