Execution Of Dataflow Graphs Research Articles

BINDING ALGORITHM FOR POWER OPTIMIZATION BASED ON NETWORK FLOW METHOD

We propose an efficient binding algorithm for power optimization in behavioral synthesis. In prior work, it has been shown that several binding problems for low-power can be formulated as multi-commodity flow problems (due to an iterative execution of data flow graph) and be solved optimally. However, since the multi-commodity flow problem is NP-hard, the application is limited to a class of small sized problems. To overcome the limitation, we address the problem of how we can effectively make use of the property of efficient flow computations in a network so that it is extensively applicable to practical designs while producing close-to-optimal results. To this end, we propose a two-step procedure, which (1) determines a feasible binding solution by partially utilizing the computation steps for finding a maximum flow of minimum cost in a network and then (2) refines it iteratively. Experiments with a set of benchmark examples show that the proposed algorithm saves the run time significantly while maintaining close-to-optimal bindings in most practical designs.

The explicit token store

This paper presents an unusually simple approach to dynamic dataflow execution, called the Explicit Token Store (ETS) architecture, and its current realization in Monsoon. The essence of dynamic dataflow execution is captured by a simple transition on state bits associated with storage local to a processor. Low-level storage management is performed by the compiler in assigning nodes to slots in an activation frame, rather than dynamically in hardware. The processor is simple, highly pipelined, and quite general. There is exactly one instruction executed for each action on the dataflow graph. Thus, the machine-oriented ETS model provides new insight into the real cost of direct execution of dataflow graphs.

Monsoon

Dataflow architectures tolerate long unpredictable communication delays and support generation and coordination of parallel activities directly in hardware, rather than assuming that program mapping will cause these issues to disappear. However, the proposed mechanisms are complex and introduce new mapping complications. This paper presents a greatly simplified approach to dataflow execution, called the explicit token store (ETS) architecture, and its current realization in Monsoon . The essence of dynamic dataflow execution is captured by a simple transition on state bits associated with storage local to a processor. Low-level storage management is performed by the compiler in assigning nodes to slots in an activation frame , rather than dynamically in hardware. The processor is simple, highly pipelined, and quite general. It may be viewed as a generalization of a fairly primitive von Neumann architecture. Although the addressing capability is restrictive, there is exactly one instruction executed for each action on the dataflow graph. Thus, the machine oriented ETS model provides new understanding of the merits and the real cost of direct execution of dataflow graphs.

A flexible model for studying the execution of dataflow programs in distributed systems

A four-stage model of the execution of dataflow graphs, the processing structure model, is presented. Each stage has its own responsibility and offers a well-defined interface to other stages. The processing structure model presents a framework for understanding, analyzing and supporting the execution of dataflow programs in a distributed system. The distributed aspects of the processing affect one or two of the four stages depending on how the stages are arranged. The model has been simulated and experimentation has been done concerning the assignment of dataflow nodes ( i.e. tasks) to computer systems. Initial results from the experimentation are presented.