Democritos is implementing a distributed virtual laboratory strongly supported by CINECA, the leading Italian super-computing center in Bologna. Democritos and CINECA have therefore installed a mini-Grid environment between Bologna and Trieste, where there is sufficient complexity to test out key middle-ware components of a Grid specifically devoted to computational material science.

The present GRID testbed is mainly based in its first implementation on Linux O.S. and comprises, beside single workstations (both IA-32 and IA-64 Intel architecture) three different Linux Clusters, two of them located at Trieste and the third one in Bologna.  The only heterogeneous architecture at the moment is the IBM SP4 supercomputer in Bologna. Figure below gives a graphical representation of our Grid.

• Application-aware scheduler
  We are implementing an application-aware scheduler in order to provide easily and seamlessly computational resources to a specific class of users with well defined requirements. We consider here a class of codes which requirements any repeated iterations of almost constant-time operations. In particular we refer to Molecular Dynamics (MD) codes where the constant-time operation is a discrete step in the time-evolution of a classical atomic system. Although Grid infrastructure implemented in our testbed provides many useful services (resource selection, work allocation functionality, information services and scheduling mechanisms) MD simulations must interpret system performance information in terms of their own requirements in order to develop performance-efficient schedules. Our approach is to define a scheduler which incorporates application specific requirements, systems characteristics and dynamic resources performance data in scheduling decisions.