Modelovanie a optimalizácia výkonnosti paralelných počítačov
Autor práce: Mgr. Dan SlováčekŠkoliteľ: Prof. Ing. Ivan Hanuliak, CSc.
Dátum obhajoby: 21.8.2013
Študijný program: 9.2.9 Aplikovaná informatika
Oponent 1: prof. Ing. Petr Dostál, CSc.
Oponent 2: prof. Ing. Jozef Sláma, CSc.
Oponent 3: doc. Ing. Ladislav Schwartz,PhD.
Anglický abstrakt:
The theses “Performance modelling of parallel computers“ deals with a problematic
ofperformance evaluation of asynchronous parallel computers (APC). The main goals were to
perform
- analysis of the development trends in the world, their actual state and perspectives in the areas parallel architectures, communication networks, criterions for performance evaluation, computing parallel models and performance evaluation methods
- queueing theory based of performance evaluation of APC including their most important components (computation, communication overheads)
- theoretical and practical performance modelling on all used forms of parallel computers (NOW, SMP and Grid) including
-- single SMP parallel computer (model M/M/m, model M/D/m)
-- network of parallel computers (network of M/M/m and M/M/1 models, network of M/D/m and M/D/1 models, comparison)
- verification of the theoretical results through
-- simulation model
-- benchmarks (SPEC tests, synthetic tests).
For real parallel computers we have been verified the achieved results on the practical examples modelling actual parallel computers as - M/M/1, M/D/1 models
- M/M/2, M/D/2 models
- M/M/4, MD/4 models
- comparison of these models.
For these modelled oparallel computers we have derived and verified analytical equations on various practical simulation experiments. The achieved and verified results could be methodically applied to other parallel computers in the future based on SMP, NOW and Grid. Key-words: parallel computer, performance evaluation methods, communication overheads, performance modelling, NOW, Grid, SMP, shared memory, distributed memory, queueing theory, simulation.
- analysis of the development trends in the world, their actual state and perspectives in the areas parallel architectures, communication networks, criterions for performance evaluation, computing parallel models and performance evaluation methods
- queueing theory based of performance evaluation of APC including their most important components (computation, communication overheads)
- theoretical and practical performance modelling on all used forms of parallel computers (NOW, SMP and Grid) including
-- single SMP parallel computer (model M/M/m, model M/D/m)
-- network of parallel computers (network of M/M/m and M/M/1 models, network of M/D/m and M/D/1 models, comparison)
- verification of the theoretical results through
-- simulation model
-- benchmarks (SPEC tests, synthetic tests).
For real parallel computers we have been verified the achieved results on the practical examples modelling actual parallel computers as - M/M/1, M/D/1 models
- M/M/2, M/D/2 models
- M/M/4, MD/4 models
- comparison of these models.
For these modelled oparallel computers we have derived and verified analytical equations on various practical simulation experiments. The achieved and verified results could be methodically applied to other parallel computers in the future based on SMP, NOW and Grid. Key-words: parallel computer, performance evaluation methods, communication overheads, performance modelling, NOW, Grid, SMP, shared memory, distributed memory, queueing theory, simulation.
Autoreferát dizertačnej práce
