|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ИНХС РАН |
||
Some estimations of performance of parallel algorithms for solving large linear systems over GF(2)статья
- Автор: Черепнев М.А.
- Журнал: Вестник Тамбовского университета. Серия: Естественные и технические науки
- Том: 15
- Номер: 4
- Год издания: 2010
- Издательство: Тамб. гос. ун-т им. Г.Р. Державина
- Местоположение издательства: Тамбов
- Первая страница: 1342
- Последняя страница: 1353
- Аннотация: This topic explains how to estimate the running time and RAM volume required by programs of Wiedemann-Coppersmith algorithm, Montgomery's algorithm, some modifications of them and new algorithm when uploading multiple compute nodes and some other details of these algorithms. In principle, each operation of any algorithm can be considered for possible use of several similar sites to reduce time of their implementation. However, if the number of arithmetic operations is fixed, the most significant reduced time - this is where time falls proportional to the growing of compute nodes' number. It will be that, for example, if you search relations in the methods of the discrete logarithm or factorization problem based on the factor databases. However, for most complex algorithms top rated time of their work T (N, n, d, c ... ; s) depend on task and used equipment parameters: N, n, d, c and the number of used compute nodes s and may behave differently on different areas of argument changing. Often when you increase s above a certain threshold s_0 (N, n, d, c ... ), depending on the task settings, time not only falls, but begins to grow again. That is because time for exchange between s computing nodes grows faster than of running time on each compute node separately reduce.
- Добавил в систему: Черепнев Михаил Алексеевич
Прикрепленные файлы
| № | Имя | Описание | Имя файла | Размер | Добавлен |
|---|---|---|---|---|---|
| 1. | Полный текст | Cherepniov11.pdf | 158,4 КБ | 4 ноября 2014 [Cherepniov] |