An asynchronous high energy efficiency map computational accelerator

Abstract

The invention discloses an asynchronous high energy efficiency map computing accelerator, which includes sequentially connected data preprocessing module, data transmission module and data processingmodule, wherein: the data preprocessing module preprocesses a given example graph G in an initialization stage, partitions large-scale graph data, and further organizes the graph G into a sub-graph data Batch Row Vector with batch_size as a size; the data transfer module transfers the sub-graph data Batch Row Vector to the on-board DDR via PCIe DMA, and then to the on-chip accelerator via AXI DMA.The data processing module is an accelerator on-chip logic, comprising a memory module and a calculation module. The memory module comprises an on-chip cache, and the computing module comprises an on-chip computing module. The invention designs the asynchronous high energy efficiency map computational accelerator, adopts the asynchronous calculation mode to accelerate the convergence speed of themap algorithm, and customizes the accelerator based on the hardware platform to reduce the power consumption and the energy consumption of the system.

Description

technical field [0001] The invention relates to graph data processing, in particular to an asynchronous high-energy-efficiency graph computing accelerator. Background technique [0002] With the gradual maturity and development of big data, cloud computing technology and the Internet industry, human life has entered the era of data explosion. Big data has the characteristics of "4V", which are large data volume (Volume), multiple data types (Variety), fast growth rate (Velocity) and low value density (Value). Graph is one of the most classic and commonly used data structures. Many data in real life are often abstracted into data with various graph structures. The vertices in the graph can represent different entities, and the edges in the graph can represent the relationship between different entities [7]. Common graph structure types of data include social networks, web graphs, transport networks, and genome analysis graphs. In addition, the scale of graph data is growin...

AI Technical Summary

Problems solved by technology

On the other hand, due to the extreme irregularity of large-scale graph data, a large amount of data communication is generated in the process of computing on traditional MapReduce and Hadoop systems, resulting in low computing efficiency.

Images