Shulou(Shulou.com)06/03 Report--

1. The means to realize parallelism in MPIMPI is process level; distributed memory system is adopted, parallel execution is realized explicitly (data allocation mode), message transmission between processes is carried out through communication, and it has good expansibility. Although MPI is suitable for all kinds of machines, its programming model is complex: it needs to analyze and divide application problems and map them to distributed process sets; it needs to solve the two main problems of large communication delay and load imbalance; it is troublesome to debug MPI programs; the reliability of MPI programs is poor, one process goes wrong, and the whole program will go wrong. 1.2 PthreadsPthreads parallelism is thread-level. It uses a shared memory system, which is valid only on POSIX systems (linux, mac OS X, Solaris, HPUX, etc.). It is a library that can connect to C programs, and the standard C++ shared memory thread library is still under development. It may be more convenient to use in C++ programs in the future. 1.3 OpenMPOpenMP is thread-level (parallel granularity); uses a shared memory system, implicitly (data allocation) to achieve parallel execution; poor scalability; precisely because of the use of shared memory distribution system, it means that it is only suitable for SMP (Symmetric Multi-Processing symmetric multiprocessing structure), DSM (Distributed Shared Memory shared memory distribution system) machines, not suitable for clusters. 1.1 OpenCL is Open Computing Language, an open computing language. Is a framework for writing programs for heterogeneous platforms that can be composed of CPU,GPU or other types of processors. OpenCL consists of a language (based on C99) for writing kernels (functions that run on OpenCL devices) and a set of API for defining and controlling the platform. OpenCL provides parallel computing mechanism based on task segmentation and data segmentation. OpenCL is similar to two other open industry standards, OpenGL and OpenAL, which are used in 3D graphics and computer audio, respectively. 1.5 GPUGPU is designed to perform complex mathematical and set calculations. There are many stream processing clusters (stream Multiprocessor) SM in a GPU, which are similar to the core of CPU. These SM are connected to the share (first-level cache) and then to the second-level cache equivalent to the interconnect switch between SM. Similar to this kind of design, it is necessary for computing graphics rendering. Hadoop is an open source version of Google's MapReduce framework. It is aimed at the linux platform. The concept is that you take a large data set and map it into many small data blocks. However, instead of sending data to each node, the dataset has been divided into hundreds or thousands of nodes through parallel file systems. Therefore, the Reduce step is to send the program to the node that already contains the data, and then write the output to the local node and save it there.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.