Data Sharing And Concurrent Access In High-through Simulation

Based on the concept of high performance computing and high throughput computing,combined with the characteristics of the application of complex simulation system,the definition of high throughput simulation is given.High throughput simulation refers to a large number of computing resources complete the large-scale simulation task with a higher efficiency of the computer simulation process in a long period of time.High throughput simulation is different from the previous simulation system.The application scope of high throughput simulation system is more extensive,and it is no longer in the specific field.And high throughput simulation,involving different users of knowledge background,the user's difference is very big.Simulation process for the user requires a large number of weakly coupled simulation work with concurrent running,and the system in the execution of the simulation work often need to instantiates a lot of reuse of simulation model,processing and analysis of mass data.In general,high throughput simulation has the characteristics of heterogeneous,high concurrency,high reliability and so on.High throughput simulation process have frequent access for I/O and simulation process is generally run in heterogeneous,dynamic and distributed system,then the access efficiency of the I/O will become a key factor for the simulation efficiency.In the simulation process,if distributed file system read or write efficiency is low,then in the whole process of simulation will be a lot of time consumption in I/O access process.Therefore,it is necessary to have a method to improve distributed file system read or write rate,and we need to ensure the high reliability,at the same time,which is the most fundamental of the research.In the process of simulation,the access to the I/O is the process of running the model for I/O access.This paper will construct a high flux simulation platform,in this platform,including simulation propulsion engines.Simulation process is a process of promoting about multiple processes,in this process,we need to a method to avoid deadlock,using the null message algorithm to avoid deadlock.In this system,the use of the memory as the file system,but owing to character of non-volatile,if only use the momery,it is difficult to guarantee reliability,and through the use of Alluxio can effectively avoid this problem.The advantages of high throughput simulation supported platform in this paper are as follows:(1)It can meet the process of high concurrent access to I/O in the process of high throughput simulation.(2)In this paper,the design of high-throughput simulation supporting platform,abandoning the traditional distributed file system based on HDFS,choice the distributed file system based on the memory.Since the rate of reading and writing of the memory medium is much higher than that of the disk,in the face of high concurrent access,the efficiency will be much higher than the traditional disk based distributed file system.(3)High throughput simulation supporting platform of the underlying storage by hierarchical structure,an upper layer of memory,the lower the HDD,do not only guarantee the I/O access efficiency and ensure the distributed file system for data storage.This paper implements a simulation supporting platform designed for high-throughput simulation,combined with the actual situation of high flux simulation process in high concurrency,high throughput system to achieve the optimization of the distributed file system,finally solve the bottleneck problem of simulation in the process of frequent access to the I/O,and enhance the simulation efficiency.The system programming with C + + Realized simulation propulsion engine and avoid deadlock of the null message algorithm,then the access to the file system was developed with the Java language,through the agency will be a combination of both the process simulation for distributed file system with high concurrency access.In this paper,the implementation of high throughput simulation support platform can effectively improve the bottleneck of access to I/O,and compared with the traditional big data simulation,there will be a higher efficiency.