ReCoNets - Goals

Goals in a Nutshell

Goals

With the objective of increasing fault-tolerance, adaptivity and flexibility on the basis of a reconfigurable network (ReCoNets), the concept of online hardware/software partitioning is proposed. Online hardware/software partitioning describes the procedure of binding functionality to free resources in the network at runtime. In order to allow for moving functionality from one node to another and execute it either on free hardware or software resources, we will introduce the concepts of task migration and task morphing. In the following Figure, a ReCoNet is given. The presented network consists of four nodes connected via point-to-point communication links. The network executes two tasks mapped on node n1 and node n4. The binding of data dependencies between the tasks are shown with dotted lines. In figure (a) a new task t3 is assigned to one of the nodes (n1) in the ReCoNet at runtime. As this assignment might violate given resource constraints (number of logic elements available in an FPGA or number of tasks assigned to a CPU), a new task binding can be demanded. Another aspect is the optimality of the resultant task binding by assigning t3 to n1. Perhaps it is desirable to distribute the workload due to the tasks in the ReCoNet. Note that a similar scenario can be induced by deassigning a task from a node. Figure (b) shows another important scenario where a link between two nodes n1 and n4 is broken. Due to this defect, it is necessary to calculate a new communication binding for the data dependency (t1,t2) which was previously routed over this link. Again a similar scenario results from reestablishing a previously broken link. Finally, in Figure (c) a node defect is depicted. In current embedded networks this would lead to a loss of functionality. But due to the separation of functionality and resources the task t2 can be bound to another node. Obviously, changing the task binding implies the recalculation of the communication binding. Besides this novel class of adaptive fault-tolerant networks, we will focus on design methodologies that allow for a convinient integration of our concepts into existing hardware/software implementations of functionality. A constraint to these design methods is the seamless integration into existing design flows.