HPC Visualization
Download as DOC, PDF0 likes268 views
The document describes a software system being developed to visually monitor the workload of cores in a high-performance manycore computer architecture. The system receives data about the state of cores in a computing system, analyzes the data, and displays it visually with remote web access. Compared to other software for visually monitoring multiprocessor systems, this system provides a visual display of processed data on the state of cores based on analysis of inter-core messages and characteristics of individual cores. The system is being developed using Microsoft Visual Studio 2008 on a 16-core Windows cluster at Polytechnic University and will aid in analyzing and monitoring complex systems and their components during different workload modes.
HPC Visualization
- 1. Michail Karpov SPbSTU, FUIT michail.karpov@gmail.com «System for visual control workload of high-performance manycores architectures» The aim is to design a software product, which includes the possibility of a visual display of multi-load systems in on-line. To develop the product chosen among MS Visual Studio 2008. Work underway on the basis of 16-core cluster running Windows HPC Server 2008 (provided by University Polytechnic University Corporation Intel) using provided by Microsoft tools and libraries and the HPC Pack HPC SDK. Product under development is a system that provides for the possibility of visual control over the workload of cores running computer system. The necessity of such a system necessitates the analysis and performance monitoring of the complex and its individual components in different modes of workloads. In addition to such controls, developed a system serves as a handy tool to enhance the effectiveness of parallel programs created to run on multiprocessor architectures. Fig. 1. The overall organization of the system being developed. Developed using Visual Studio 2008 program receives data about the state of the nuclei that make up the computing system, analyzes them, and choosing the most convenient for the end user's visual presentation, displays the received data to the screen with remote web access. Compared with other software designed for visual inspection of multiprocessor systems, this development provides a visual display of the processed data on the state of the nuclei of the complex, based both on the analysis of messages sent between cores, as well as on the characteristics of each core separately (at the moment query processing). Scientific publications: • "The system for visual inspection of high-load multi-core architectures, periodicals scientific peer- reviewed journal" Scientific and technical statements SPbSTU", M.Karpov, 2009 • "Visualization of parallel programming" Science and innovation in technical universities: Proceedings of the National Forum of Students and Young Scientists, M.Karpov, 2008 • "Analysis and control performance of multiprocessor architectures and their individual components", abstracts for participation in Science Week XXXVIII SPbSTU, M.Karpov, 2010 • "The system for visual inspection of high-load multi-core architectures". Proceedings of the III session of the scientific school-workshop for young scientists and specialists "Technology High Performance Computing and computer simulation", M.Karpov, 2009
- 2. Michail Karpov SPbSTU, FUIT michail.karpov@gmail.com «System for visual control workload of high-performance manycores architectures» 1. How much load the program has on the system? (operates on a single core? what resources are used? how affects the performance of the cluster? how affects the accuracy of the estimate congestion? A: Works on the head node, the bulk of interference can occur due to use of system resources for rendering results - will display them on a remote machine. 2. What are the characteristics of the system displays the program? A: Free memory, memory allocation processes running on the server, and provides the ability to monitor and change the status of implementation activities and tasks on the cluster. 3. How often they update the information on the state system? A: Information about the core (number of nodes and memory on them) are updated separately (at least) from the information load of the nuclei (in seconds). 4. But what if so many cores? A: For this purpose the concept of multiple entry points into the system, and display only the required information at this time and detail with a further approximation. 5. What problems are simulated on this system? A: Basically - to work with graphic images. 6. On which machines of the cluster type you worked and what problems you solved? A: MSU - Blue Gene / P - 23.8 TFlops Linpack (378 place in the world Top500) - Multiplication of large matrices, working with graphics. Hardware-software complex T-Forge Mini on the basis of eight dual-core AMD Opteron processor and operating system Microsoft Windows Compute Cluster Server 2003 at UNN. Also - a 16-nuclear cluster FUIT running Windows HPC Server. 7. What are the approximate performance of supercomputers? A: Supercomputer "Lomonosov" (12 line in the Top500): peak performance 420Tflops, 35,776 processor cores, 576GB RAM 56, the power consumption calculator 1.5 MW, 252 sq ft footprint, interconnect QDR Infiniband (40 Gbits / c), the amount of storage up to 1350TB. 8. On what basis is the metaphor of the molecule is more convenient? Why she was not offered in any of the existing projects? A: Based on research, V.L. Averbukh and his research in this area. Existing projects are mainly designed for systems with small (less than 500) the number of processor cores. 9. What is HPC Server 2008? What he especially? A: Cluster version of the OS designed to address high-tech tasks that require computing cluster. Provides a scalable management tools cluster, high technology NetworkDirect RDMA, service-oriented architecture (SOA) job scheduler. Windows HPC Server 2008 enables you to achieve the computing speed by an average of 25-30% more than the previous version of the product. 10. Why do I need to display congestion? A: To analyze which of the cores are idle, to distribute the work evenly between them for speed and system stability. 11. Which, basically, the problem be solved? A: Displaying a large (over 1000) number of cores and objectives, a parallel implementation of the program. 12. Can you work remotely? A: Yes, for the moment the main work is done in this direction. 13. What characteristics can still be displayed? A: Planned: air-moving, pressure, free volume, humidity level, speed of communication channels. 14. Why draft a perspective? A: Supported by a grant from St. Petersburg, awarded a diploma from Microsoft, the project invited the cooperation of the director of a major information-processing complex Polytechnic University, and Chair of "Mathematical and software high-performance computing (MiPOVV). In the long term - a direct part in the program of the Russian Academy of Sciences, Synterra and Hewlett-Packard "University Cluster". Main scientific rewards: • Grant of the Government of St. Petersburg: for diploma project "System for visual inspection of high- load multi-core architectures“, 2009 • Diploma: 1 st place in the Conference-competition "Technologies of Microsoft in the theory and practice of programming", representative of Microsoft, St. Petersburg
- 3. • Diploma: "For the scientific results of fundamental and applied exploratory research", FUIT STU-up to the All-Russian Scientific Conference «XXXVII Science Week STU"