Font Size: a A A

A Study Of Metis Framework Based On Embeded Multiprocessor

Posted on:2015-11-04Degree:MasterType:Thesis
Country:ChinaCandidate:S X ZhengFull Text:PDF
GTID:2308330464968620Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Parallel computing on embedded platform has become increasingly popular. The single-core processor can not meet the needs of the application longer.It’s widely used for multi-core processors to promote the development of parallel computing embedded. Many technologies of parallel computing,such as MPI, Open MP and so on. A lots of software framework based on the Map Reduce programming model flourish in the computer field, such as Spark, Hadoop, Metis and Phoenix and so on. Spark and Hadoop are generally used for PC cluster. It is not used for a cluster on embedded platform. Metis is a programming framework based on the Map Reduce, which is a improvement of Phoenix. It spend less resources than Hadoop. So, Metis and Phoenid are suitable for parallel computing on embedded platforms. Metis and Phoenix are programming models based on shared memory.So they are only support communication between the nucleus inside the node.In this paper, Metis framework based on the Map Reduce programming model has been studied.In order to make Metis framework work on the cluster node with an embedded platform. Metis are designed as a core, which mainly execute task based on Map Reduce. And MPI is used to communicate between nodes. This framework called MPI+Metis.Assess differences in test MPI+Metis and MPI+Phoenix’s framework by testing the Metis standard set.Finally, in the last of paper, it assesses the embedded cluster node load condition of node by collecting the CPU and memory usage.Finally,the application of Word Count and Linear Regression are test the MPI+Metis and MPI+Phoenix framework, which indicate the framework based on Metis is more effective than Phoenix in data procedure.Test the CPU usage and memory usage on cluster nodes on multiple embedded indicate that the larger the amount of data applications, the time of CPU usage and memory usage at peak are longer, but a cluster of multiple nodes can reduce the system load.
Keywords/Search Tags:Embedded, Map Reduce, MPI+Metis
PDF Full Text Request
Related items