|
Modern high performance computing resources allow to execute a set of experiments
on modelling of difficult technological processes and they show impressing results
[7]. There is the percentage statistics usage of supercomputers computing potential
in various projects and industries in a figure 1. Some of modern researches, which
are executing on high perfomance computing resources, will play the important role
in the future, for example, analysis and modelling of global warming, studying of
processes which occur during earthquakes, studying of ozone gaps. Development of
new technologies improvements a humanity lifes, for example, development of the
powerful antenna for mobile phones, studying of electrical nanomaterial properties.
Most widely supercomputers are used in aerospace (for example, development of a
new generation of supersonic transport).
Figure 1 — Statistics of supercomputers usage for various branches [9].
Scientists and engineers who use computing resources for difficult calculations,
meet with difficulties in design of tasks for performance in the environment of
Grid. It is linked to serious distinctions between modern high perfomance platforms
(for example, different ways of access and procedures of authentification, distinction
in the server architectures) [6]. For overcoming of the given limitation systems
of automation modelling in the environment of Grid are used. In the future the given
systems will allow to use the distributed high perfomance computing systems everywhere.
Efficiency and correctness of data exchange performance in networks Grid are one
of the main metrics of correct complex network functioning. For reception of the
estimated information about various Grid components benchmarks are used. Results
of the given programs very often are criterion for a choice of the equipment. Successful
passing of a number benchmarks can be the certificate of system stability [3]. The
subsystem of testing and optimisation will allow to reveal the majority of defects
in data transfer and search of possibilities for improvement of a network exchange
in distributed high perfomance computing systems. The given subsystem is developed
for systems of automation modelling (for example, Science Experimental Grid Laboratory)
in the conditions of the virtual organisation which possesses high perfomance computing
resources [7,8].
One of the primary goals of testing subsystem development of data exchange mechanisms
is reception of the answer to a question «Really all necessary data for the started
tasks are transferred so fast how it allows the given distributed high perfomance
computing system?». Developed subsystem allows to fix character and the main features
of solved tasks. Further it presumes to execute any improvements at construction
of the distributed high-efficiency computing systems on the basis of the revealed
processes and legitimacies.
The purpose of master work is development of a universal subsystem of data transfer
testing in the distributed high perfomance computing systems which will allow to
perform complex testing for collection of the main statistical information for the
subsequent improvement for communications between Grid components. The information
about network exchange can receive not only administrator, but also the user.
It is necessary to solve a number of tasks:
- Check of net points availability for "peer-to-peer" exchanges.
- Check of correctness and efficiency of used data exchange mechanisms.
- The organisation of load and stress testing for definition of maximum loads on system.
Implementation of the given task will allow to reveal not only disadvantages of
a communication network, but also to select sites on which the information can be
transferred with a high speed.
- The organisation of comparative testing for definition of used data exchange mechanisms.
Research of advantages and disadvantages for various data exchange mechanisms will
allow to select best of them further.
It is necessary to research all possible optimisation variants of considered data
exchange mechanisms. One of results is the optimal parametres set for customisation
of the data exchange mechanism. However, change of functioning customisations in
most cases demands human interference, therefore also it is necessary to carry out
research of automatic optimisation for considered data exchange mechanisms.
Grid calculations have arisen as the new and important area different from traditional
distributed computing and linked to necessity of large-scale resources and for allocation
of high perfomance data processing. Carrying out of difficult calculations in the
virtual organisations (associations of single users, institutes and resources) is
new technology and demands the constant control [2].
Many difficult systems are under construction of ready components, including high-speed
networks. A success basis is realisation of a correct components choice of such
systems on the basis of complex estimated testing [1].
Testing and optimisation subsystem will be a new look at discipline of the distributed
high perfomance systems testing. The given subsystem will execute complex analysis
of efficiency for the used data exchange mechanism that will allow to make its automatic
customisation further. At addition of new resources in network Grid they will be
automatically "picked up" and to stand in line on testing.
Ultimate goal of master's work is creation of testing and optimisation subsystem
for data exchange mechanisms in distributed high perfomance computing system. This
subsystem will be realised in Java as the component of project SEGL of Shtuttgart
University. Integration into the given project will allow to execute the main debugging
and the systems analysis of the received results for the real high perfomance computing
system.
The modern distributed high perfomance computing systems are difficult for optimisation
(for example, network productivity can worsen with physical or logical distance
between sites). Therefore it is difficult to define optimal properties and consequently
and somehow to improve the data transfer mechanism. The subsystem of the testing
will allow to automate process of testing and optimisation of used data exchange
mechanisms in distributed high perfomance computing systems, allowing to overcome
their some limitations.
- Эйсымонт Л.К. Оценочное тестирование высокопроизводительных систем: цели, методы,
результаты и выводы / Центр независимого межведомственного тестирования суперкомпьютерных
систем.
[ Source:
www.nicevt.ru/Repository/file/article-03.pdf ]
- Фостер Ян, Кессельман Карл, Тьюке Стив. АНАТОМИЯ ГРИД.
[ Source:
www.gridclub.ru/library/publication.2004-11-29.7104738919/publ_file/ ]
- Benchmark (computing) / Wikipedia, the free encyclopedia.
[ Source:
www.en.wikipedia.org/wiki/Benchmark_(computing) ]
- Currle-Linde N., Kuster U., Resch M., Risio B. Science Experimental Grid Laboratory
(SEGL) Dynamical Parameter Study in Distributed Systems / Parallel Computing: Current
& Future Issues of High-End Computing, John von Neumann Institute for Computing,
2006, Julich
[ Source:
www.fz-juelich.de/nic-series/volume33/049.pdf ]
- Currle-Linde N. , Resch M., Kielmann T., Wrzesinska G. REDESIGNING THE SEGL PROBLEM
SOLVING ENVIRONMENT: A CASE STUDY OF USING MEDIATOR COMPONENTS.
[ Source:
www.cs.vu.nl/~kielmann/papers/segl.pdf ]
- Интернет-портал по грид технологиям. Платформа UNICORE
[ Source:
www.gridclub.ru/software/unicore.html ]
- SEGL — Science Experimental Grid Laboratory. Grid Management System
[ Source: www.segl.hlrs.de
]
- Коваленко В.Н., Корягин Д.А. Организация ресурсов грид, ИПМ им. М.В.Келдыша РАН,
Москва 2004.
[ Source:
www.keldysh.ru/papers/2004/prep63/prep2004_63.html ]
- Сайт компании NEC.
[ Source: www.nec.com ]
- Vraalsen Fredrik, Aydt Ruth A., Mendes Celso L., Reed Daniel A. Performance Contracts: Predicting and Monitoring Grid Application Behavior
[ Source: www.springerlink.com/content/yx4pmjr9wxf91hpt/ ]
|
