Mpich-based Parallel Computing System, Load Balancing Technology

According to the expansion of super computing requirements,high performance computing has helped people in data processing areas such as:data mining applications,image processing business,gene sequencing comparing.However,considering the rapid expansion of the scientific computing scale,the traditional serial calculation has been unable to meet the requirements,thus the concept of parallel computing has been carried out.Parallel computing is in contrast to the serial calculation,it distributes the processes to several independent nodes,which are scheduled by independent operating system,and have independent CPU and memory resources;the exchanges of information between nodes are through message transmission.There are some important improvements on computer system architecture in the recent 50 years although there was no revolutionary changes except the data flow computer which gradual improvement toward practical application.The computers have been put into applications are all based on Von Neumann prototype,and improve the means and capabilities of parallel processing on this basis.MPI is a kind of message passing interface which is published on May,1994,it provides the standard library for message passing procedure.MPI defines the Interface library by language independent form,and provides the binding with C and FORTRAN language.It is the most reliable platform for high efficiency and ultra-large-scale parallel computing.This paper from research task allocation,load balancing starting,improved in the MPICH task allocation shortcoming,and constructed one based on the MPICH load balancing model.Model using the concept of relative processing speed,the handlocalhost capacity of each node were quantified, scheduling nodes allocate task according to the capacity of each node weights in order to achieve the load balancing system as a whole.In order to use the extensively distributed idle calculation resources on the Internet,there are already a number of distributed computing projects have been setup,for example:SETI@Home, GIMPS and so on.These projects have confirmed the feasibility of such calculation model. These resources are highly unstable and cannot be trusted by users,bandwidth is limited and full of tremendous delay.In order to compensate for the deficiencies of parallel computer systems, load balancing technology is the concern of people and rapidly developed of. In this paper,the critical algorithm of MCP is improved,and advanced a algorithm of TPG. Start with the idea in theory,an analysis of differences in schedulocalhost and dispatching of the different results,then use the algorithm for parallel computing environments and modify the key path algorithm and analysis were compared.