A Data Flow Graph Congestion Detection Method Based on Execution Efficiency Gradient Prediction

Abstract

The present invention discloses a data flow diagram congestion detection method based on execution efficiency gradient prediction. The method comprises: setting a timer and a instruction counter at each node in a data flow diagram separately; setting a management node at the data flow diagram; setting a first information record table at the management node, wherein an ID, an execution rate v, an execution rate change rate s, a predicted execution rate vn and a queue mark k of each node are recorded in the first information record table; setting a second information record table at each node; performing congestion detection on each node and calculating the execution rate v, the execution rate change rate s and the predicted execution rate vn of the node; sending the detected execution rate v, execution rate change rate s and the predicted execution rate vn and the ID and queue mark k of the node to the management node so as for the management node to update the first information record table and send congestion information of an adjacent node to the node and for the node to update the second information record table accordingly.

Description

technical field [0001] The invention relates to the technical field of data flow architecture, in particular to a data flow graph congestion detection method based on execution efficiency gradient prediction. Background technique [0002] With the advent of the era of big data, the amount and scale of data that computers need to process is increasing. The data flow architecture has attracted people's attention due to its high instruction parallelism and low memory access frequency. In a dataflow architecture, program code can be transformed into a dataflow graph. A data flow graph is a directed graph in which each node represents a piece of code in a program and directed lines represent the flow of data between nodes. In the data flow graph, the execution rates of different nodes are different, which will cause the congestion of the data flow graph. In order to alleviate the congestion of the data flow graph, methods such as multi-context and key node replication can be us...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that there are too many nodes to be detected by special nodes, and the demand for network bandwidth is large.

In this way, the acceleration method can be fully utilized, so that the acceleration resources can be allocated reasonably; if only the congestion of itself is detected, it is likely to cause excessive competition for acceleration resources, increase bandwidth requirements, and reduce execution efficiency

Images