News 
-
National Supercomputing Forum (NSCF-2015) was held on November 24th - 27th, 2015 in Russia, Pereslavl, the Program Systems Institute of RAS.
Presentations
Particularity of development of program with irregular grids through the use of DVM-system.
Particularity of development of program with irregular grids through the use of DVM-system.
This paper discusses the possibility of using the DVM-system to develop applications that use irregular grids. Authors analyzes the limitations of DVMH-model, which reduce performance of these applications. The proposed extensions of this model should greatly simplify the development of parallel programs with irregular grids that can be efficiently executed on heterogeneous computational clusters which use not only universal multi-core processors but also accelerators (GPUs or Intel Xeon Phi coprocessors).
Presented by M.N. Pritula.
-
Russian Supercomputing Days 2015 was held on Semptember 28th - 29th, 2015 in Russia, Moscow, hotel Holiday Inn Moscow - Sokolniki.
Presentations
Automated parallelization of sequential C-programs on the example of two applications from the field of laser material processing
Automated parallelization of sequential C-programs on the example of two applications from the field of laser material processing
It is important to understand the information structure of programs for their parallelization. This helps to realize which kind of transformations may be necessary, and which parts of source code can be executed in parallel. Systems for automated analysis and transformation of programs may be useful to explore the structure of programs and to improve the performance of parallelization within a reasonable period of time. The paper proposes an approach to developing of such kind of systems. The process of program transformation is split into a set of basic operations. These operations are performed automatically in the order which is determined by the user. The offered approach has been successfully applied to parallelize two applications from the field of laser material processing.
Presented by N.A. Kataev.
Methods of dynamic tuning of DVMH programs on clusters with accelerators
Methods of dynamic tuning of DVMH programs on clusters with accelerators
DVM system is intended for development of parallel programs of scientific-technical calculations in C-DVMH and Fortran-DVMH languages. These languages use single model of parallel programming (DVMH model) and they are extensions of standard C and FORTRAN languages by the specifications of parallelism issued in the form of directives to the compiler. DVMH model allows to create effective parallel programs for heterogeneous computing clusters with accelerators. Using DVMH model the programmer doesn’t use explicit copy operations of the data located in the memory of central processor (CPU) or accelerators. For program fragments (regions) which can be executed on accelerators, he specifies input and output data, and also those data which are updated or used out of regions. It allows to select dynamically the devices on which the region will be executed, to distribute job between devices according to their productivity, to execute repeatedly regions for selection of optimum configuration. The influence of the listed methods on execution efficiency of some tests (from NAS NPB benchmarks) and real applications is shown in the article.
Presented by V.A. Bakhtin.
-
Scientific service on the Internet 2015 was held on Semptember 21st - 26th, 2015 in Russia, Novorossiysk, Abrau-Durso, a boarding house 'Sailor'.
Presentations
Using the Internet for study of parallel programming
Using the Internet for study of parallel programming
Using of Internet is a promising method of teaching that can increase the effectiveness of intramural and extramural attendance. Network technologies provide an interactive communication with the teacher, a regular monitoring of the educational process, an access to Internet resources containing real-time information on relevant topics. During the study of supercomputer technologies it is necessary to have access to high performance systems that are physically located in supercomputer centers and these systems can be accessed only remote. It is vital to realize that Internet resources provide access to a wide range of unstructured data sources, the accuracy of which is not guaranteed. The paper discusses the use of Internet in the process of learning supercomputer technologies at the Faculty of Computational Mathematics and Cybernetics of Lomonosov Moscow State University. Students acquire the necessary skills of working with remote computing systems, they get an overview of the rules of working on supercomputers, they study resource management systems to run their programs. The course includes an introduction to the following parallel programming technologies: MPI, OpenMP, OpenACC and DVMH. The authors have developed DVM-system website which contains a detailed guidance on the use of this system, sample programs, publications and reports. The control system of the learning process is organized on the basis of a project management web application.
Presented by N.A. Kataev.
-
Parallel computational technologies (PCT) 2015 was held on March 30th - April 3rd, 2015 in Russia, Ekaterinburg, Ural Federal University, Institute of Mathematics and Mechanics, Ural Branch of the RAS.
Presentations
Automation of serial Fortran-program transformation, nessesary for its efficient parallelization by SAPFOR system
Automation of serial Fortran-program transformation, nessesary for its efficient parallelization by SAPFOR system
Automatic mapping of serial programs on computing systems with distributed memory may require pre-transformation of the programs, oriented on the class of the systems. SAPFOR system usage for parallelization of application programs has allowed to identify the transformations, the implementation of which can be automated. The article presents the transformations that increases the possibility of effective parallelization of programs by eliminating the causes preventing to parallelization of loops. The implementation of these transformations has allowed to automate the obtaining of serial implementation, effeciently mapped on modern clusters by automatically parallelizing system compiler, for the problem of hydrodynamics.
Presented by N.A. Kataev.
Parallelization of NAS NPB3.3.1 tests on Fortran-DVMH for Intel Xeon Phi coproccessor
Parallelization of NAS NPB3.3.1 tests on Fortran-DVMH for Intel Xeon Phi coproccessor
The article analyzes the efficiency of the implementation of NAS benchmarks from NPB 3.3.1 package (EP, MG, BT, SP, LU) on cluster nodes with different architectures using multi-cores processors, NVidia graphics accelerators and Intel coprocessors. Characteristics of the tests developed in high-level Fortran-DVMH language (hereafter referred to as FDVMH), and their implementations in other languages are compared. The effect of different optimization methods for FDVMH NAS benchmarks necessary for their effective work in Intel Xeon Phi coprocessor is researched. The results of the simultaneous using of all cores in CPU, GPU and Intel Xeon Phi coprocessor are presented.
Presented by V.A. Bakhtin.
-
Performance benchmarks from NAS NPB 3.3 had been parallelized using Fortran-DVMH language and were executed on graphics accelerators NVIDIA, on 6-cores processor Intel and on 60-cores Intel Xeon Phi.