From the fron processor, scheduler module get a set of interpretive frames, each one is a somehow independent computation module. Then we need to manage their execution in the schduler and generated the results for the front processor.
According to the interpretive frames, it maybe just a directly interpreted frame, and we will execute the frame locally. In some cases, the interpretive frames may invoke a precompiled computational function, sequential or parallel. Then we need to call the function in the computational library and run to generate the results. This will need the dynamic linking mechanism to execute because what we are provided is an interactive interpreter.
Because the interpretive frames provided from the front processot are somehow independent objects. According to the availability of computing ability, the computation functions available, and the purpose of minimal execution time for the interpretive frames, we need the migrate some tasks to some other machines to finish the computation.
Task migration is the ability to suspend the execution of a sequential program on one machine and restart it, in mid-execution, on a different machine. In the context of parallel programming, this sequential program is part of a larger parallel program. Several existing systems are already capable of migrating tasks during execution and many aspects of task migration have already been discussed, such as PVM and MPI. But they are based on the execution codes and they will meet problems when a more ``exotic'' and much longer-term research effort may center around heterogeneous task migration.
In our scheduler module, we will have a set of interpretive frames from front processor, which are in some intermidiate form, not the execution codes. This makes us comparetively easy to migrate a certain tasks from one computer to another in a heterogeneous way. These interpretive frames will form a common runtime execution model for data processing of scientific computing among the computers. (Refer to the Java bytecode)
From the mechanism of task migration, we can send some tasks to some registerd remote system to finish the computation, and then concate the results in the scheduler for the front processor. This mechanism will make the system have the ability of INTERNET computing, making use of the global computing resource available.