Research On High Performance Parallel Distributed Embedded Cluster Construction And Application

With the advent of the mobile Internet ear,people want embedded applications can become more intelligent.It makes embedded applications need more and more computing power.Therefore,it will challenge the computing performance of the embedded system,and the embedded cluster system emerged as the times require.At present,the research of embedded cluster is still in the initial stage and there are still some shortcomings in many ways.First,computing resources of the embedded node are limited,so that the overall performance of the cluster is not high.Second,the expansion of the cluster is not good,when the node increased,the growth of the cluster performance decreased significantly.Third,many studies have just stayed in the performance testing phase,and lack of application research.To solve the problems,in this paper,we focuses on the construction and application of a high performance,high scalability,parallel and distributed embedded cluster as the center,and carry out the following research work.Firstly,we built the high performance parallel and distributed embedded computing cluster and tested its performance.In this paper,we chose the embedded processor with internal parallel structure,constructed a fully functional embedded cluster with two levels parallel structure,and provided reasonable software architecture for the cluster.Experimental results show that the computing power of the embedded cluster has exceeded the 10GFLOPS(100 million floating-point operations per second)and the two levels parallel structure brought 10 times speed up to the cluster.And then,through the analysis of the experimental results,this paper gives some methods and suggestions to improve the performance of embedded cluster.Secondly,to solve a problem that Canny edge detection algorithm can not be calculated in real time in embedded system,the embedded cluster is applied to detect image edge and this paper presents PDEC-Canny algorithm which can adapt to different real-time scene.In the real-time demand lower scenario,experimental result show that the algorithm is faster 4.9 times than traditional Canny algorithm on the embedded cluster which have 4 nodes,when detecting the edge of the Lena image which size is 512"512.In the real-time demand higher scenario,the experimental result is 17.71 times.Therefore,the new algorithms based on embedded cluster has better real-time.Finally,according to the cluster improvement methods and suggestions,we added an embedded CPU and GPU heterogeneous computing node for the cluster.In order to illustrate the effectiveness of the improvement,we improved Canny algorithm and proposed ECG-Canny algorithm on embedded CPU and GPU structure.Experiments show that using ECG-Canny to detect the edge of Lena image which size is 512×512 only need 3 ms.Therefore,the algorithm greatly improves the real-time performance of the edge detection application in embedded system.It also reflects that adding the embedded CPU and GPU heterogeneous nodes can effectively improve the computing power of embedded cluster.