Resultados Benchmark » History » Version 13

Thiago Paiva, 19 Apr 2013 13:46

1 1 Raphael Cóbe
h1. Resultados Benchmark
2 1 Raphael Cóbe
3 11 Thiago Paiva
4 12 Thiago Paiva
h2. Nomenclatura
5 12 Thiago Paiva
6 12 Thiago Paiva
7 12 Thiago Paiva
8 11 Thiago Paiva
h2. Instruções para teste de "single"
9 1 Raphael Cóbe
10 12 Thiago Paiva
* criar diretório e acessá-lo para receber informações do teste. O nome do diretório deve obedecer a nomenclatura estabelecida.
11 11 Thiago Paiva
<pre>
12 1 Raphael Cóbe
~$ mkdir <nome>
13 12 Thiago Paiva
~$ cd <nome>
14 11 Thiago Paiva
</pre>
15 11 Thiago Paiva
16 1 Raphael Cóbe
* copiar os arquivos necessários para o diretório recém criado
17 11 Thiago Paiva
<pre>
18 13 Thiago Paiva
~$ cp -r /opt/hpl/* .
19 11 Thiago Paiva
</pre>
20 1 Raphael Cóbe
21 11 Thiago Paiva
* modificar o arquivo "HPL.template" nos quesitos P e Q
22 11 Thiago Paiva
<pre>
23 12 Thiago Paiva
~$ vim hpl-files/HPL.template
24 1 Raphael Cóbe
</pre>
25 11 Thiago Paiva
26 11 Thiago Paiva
* editar o script "generateFile.pl" e alterar os parâmetros Ns e NBs
27 11 Thiago Paiva
<pre>
28 12 Thiago Paiva
~$ vim hpl-files/generateFile.pl
29 11 Thiago Paiva
</pre>
30 11 Thiago Paiva
31 11 Thiago Paiva
* executar script "generateFile.pl" para gerar arquivos de entrada
32 11 Thiago Paiva
<pre>
33 12 Thiago Paiva
$ perl hpl-files/genereteFile.pl
34 11 Thiago Paiva
</pre>
35 11 Thiago Paiva
36 11 Thiago Paiva
* modificar "runme-single.sh" no quesito número de cores (NUMBER_CORES)
37 11 Thiago Paiva
<pre>
38 11 Thiago Paiva
~$ vim hpl/runme-single.sh
39 11 Thiago Paiva
</pre>
40 11 Thiago Paiva
41 11 Thiago Paiva
* executar runme-single.sh
42 11 Thiago Paiva
<pre>
43 11 Thiago Paiva
~$ hpl/runme-single.sh
44 11 Thiago Paiva
</pre>
45 11 Thiago Paiva
46 11 Thiago Paiva
47 11 Thiago Paiva
48 11 Thiago Paiva
49 9 Raphael Cóbe
h2. 1 Máquina Real com HT desabilitado;
50 2 Raphael Cóbe
51 1 Raphael Cóbe
* Máquina Educloud05:
52 2 Raphael Cóbe
53 2 Raphael Cóbe
h3. Ambiente:
54 2 Raphael Cóbe
55 2 Raphael Cóbe
* 8 Processadores de 2.67 GHz;
56 2 Raphael Cóbe
* 6 GB de Ram;
57 2 Raphael Cóbe
* 90% de Uso de memória:
58 4 Raphael Cóbe
* QxP: 1x8;
59 1 Raphael Cóbe
60 4 Raphael Cóbe
h3. Resultados:
61 2 Raphael Cóbe
62 4 Raphael Cóbe
* Valor de NB ótimo: 120 
63 1 Raphael Cóbe
* Número de Operações de Ponto Flutuante por Segundo: 7.66826e+01 GFlops;
64 4 Raphael Cóbe
* Gráfico de Tempo vs NB vs Número de Operações:
65 4 Raphael Cóbe
!https://ncc.unesp.br/its/attachments/226/NBvsTimevsGflops.png!
66 4 Raphael Cóbe
67 4 Raphael Cóbe
68 4 Raphael Cóbe
69 9 Raphael Cóbe
h2. 1 Máquina Real com HT habilitado
70 4 Raphael Cóbe
71 4 Raphael Cóbe
* Máquina Educloud06:
72 4 Raphael Cóbe
73 4 Raphael Cóbe
h3. Ambiente:
74 4 Raphael Cóbe
75 4 Raphael Cóbe
* 16 Processadores de 2.67 GHz;
76 4 Raphael Cóbe
* 6 GB de Ram;
77 4 Raphael Cóbe
* 90% de Uso de memória:
78 4 Raphael Cóbe
* QxP: 1x16;
79 4 Raphael Cóbe
 
80 4 Raphael Cóbe
h3. Resultados:
81 4 Raphael Cóbe
82 4 Raphael Cóbe
* Valor de NB ótimo: 96 
83 4 Raphael Cóbe
* Número de Operações de Ponto Flutuante por Segundo: 6.71501e+01 GFlops;
84 2 Raphael Cóbe
* Gráfico de Tempo vs NB vs Número de Operações:
85 5 Raphael Cóbe
!https://ncc.unesp.br/its/attachments/227/NBvsTimevsGflops.png!
86 6 Raphael Cóbe
87 6 Raphael Cóbe
h4. Comparação
88 6 Raphael Cóbe
* Gráfico Comparativo entre Máquina com HT e sem ht habilitado:
89 7 Raphael Cóbe
!https://ncc.unesp.br/its/attachments/228/NBvsTimevsGflops-Comparison.png!
90 8 Raphael Cóbe
91 1 Raphael Cóbe
h4. Discussão
92 1 Raphael Cóbe
* O uso do HT claramente piora o resultado;
93 9 Raphael Cóbe
94 9 Raphael Cóbe
95 9 Raphael Cóbe
h2. 1 Máquina Real com HT habilitado (benchmark 2)
96 9 Raphael Cóbe
97 9 Raphael Cóbe
* Máquina Educloud06:
98 9 Raphael Cóbe
99 9 Raphael Cóbe
h3. Ambiente:
100 9 Raphael Cóbe
101 9 Raphael Cóbe
* 16 Processadores de 2.67 GHz;
102 9 Raphael Cóbe
* 6 GB de Ram;
103 9 Raphael Cóbe
* 90% de Uso de memória:
104 9 Raphael Cóbe
* **QxP: 2x8**;
105 9 Raphael Cóbe
 
106 9 Raphael Cóbe
h3. Resultados:
107 9 Raphael Cóbe
108 9 Raphael Cóbe
* Valor de NB ótimo: 104 
109 9 Raphael Cóbe
* Número de Operações de Ponto Flutuante por Segundo: 6.80169e+01 GFlops;
110 9 Raphael Cóbe
* Gráfico de Tempo vs NB vs Número de Operações:
111 9 Raphael Cóbe
!https://ncc.unesp.br/its/attachments/229/NBvsTimevsGflops-2x8.png!
112 9 Raphael Cóbe
113 9 Raphael Cóbe
h4. Comparação
114 9 Raphael Cóbe
* Gráfico Comparativo entre Máquina com razão entre QxP mais 'quadrado'  e 'linear':
115 9 Raphael Cóbe
!https://ncc.unesp.br/its/attachments/230/NBvsTimevsGflops-Comparison1x16HTvs2x8HT.png!
116 9 Raphael Cóbe
117 9 Raphael Cóbe
h4. Discussão
118 10 Raphael Cóbe
* O uso de matrizes QxP mais quadradas causa impacto no resultado:
119 10 Raphael Cóbe
** Tempos de execução foram menores;
120 10 Raphael Cóbe
** Quantidade de operações executadas também foram maiores;