Panel “Making real large-scale grids for real money-making users: why, how and when?” August 2005 Achim Streit Forschungszentrum Jülich in der Helmholtz-Gesellschaft
2Forschungszentrum Jülich Large? ... in number of participating sites virtual organization management ... in compute capability resource management ... in storage capacity data and replica management, data access and transfer ... in geographical expansion network latency, bandwidth, availability ... in amount of users authentication/authorization, accounting/billing ... in middleware platforms interoperability
3Forschungszentrum Jülich Important aspects in large-scale Grids easy and straightforward installation and configuration of Grid middleware platform full interoperability of Grid middleware platforms, if more than one is used seamless integration of new sites and resources in the Grid infrastructure scalability and redundancy of major components and services for e.g. information, scheduling, accounting,... smooth and fast integration of new users at all sites common rules, policies, and methods for e.g. security mechanisms, directory & file naming conventions,... powerful monitoring system for jobs and resources
4Forschungszentrum Jülich DEISA’s approach UNICORE as Grid middleware platform DEISA primer for new site administrators with all information about naming conventions, operational methods, configuration files, hostnames and ports of core services,... dedicated network from NRENs and GEANT between the DEISA sites for job migration and data transfer GPFS as storage and access to data LDAP for user data and integration in UNICORE User DataBase (UUDB) LL-MC and LSF for resource management, co-allocation, job re-routing, and migration Ganglia, Inca, and MDS2 for automated testing and monitoring