[en] The authors have prototyped and analyzed design of a novel approach for the high throughput computing - a core element for the emerging HENP computational grid. Independent event processing in HENP is well suited for computing in parallel. The prototype facilitates use of inexpensive mass-market components by providing fault tolerant resilience (instead of the expensive total system reliability) via highly scalable management components. The ability to handle both hardware and software failures on a large dedicated HENP facility limits the need for user intervention. A robust data management is especially important in HENP computing since large data-flows occur before and/or after each processing task. The architecture of our active object coordination schema implements a multi-level hierarchical agent model. It provides fault tolerance by splitting a large overall task into independent atomic processes, performed by lower level agents, synchronizing each other via a local database. Necessary control functions performed by higher level agents interact with the same database thus managing distributed data production. The system has been tested in production environment for simulations in the STAR experiment at RHIC. Authors' architectural prototype controlled processes on more than a hundred processors at a time and has run for extended periods of time. Twenty terabytes of simulated data have been produced. The generic nature of their two level architectural solution for the fault tolerance in distributed environment has been demonstrated by its successful test for the grid file replication services between BNL and LBNL