|
|
沙发

楼主 |
发表于 2008-11-28 11:36:21
|
只看该作者
Re: 关于SGIO的疑问!急用!各位!
有时间找了点东西看,不知道有用没
Suppose that you test for throughput improvements by changing only the BUFNO= value, as illustrated here:
[code:4eozqpif]options bufno=n;
data sample;
do i=1 to 100000000;
output;
end;
run;[/code:4eozqpif]
If you submit this DATA step five times using BUFNO= option values that range from 1 to10,000, you obtain the following
time results, measured in seconds.
[code:4eozqpif]SAS Run Run 1 Run 2 Run 3 Run 4 Run 5 Average
NOSGIO BUFNO=1 (default) 40.68 40.81 37.56 43.23 40.74 40.60
SGIO BUFNO=3 57.27 59.10 57.99 58.10 59.71 58.43
SGIO BUFNO=20 31.01 35.31 31.03 31.04 30.73 31.82
SGIO BUFNO=50 29.35 30.06 30.26 30.23 30.02 29.98
SGIO BUFNO=100 29.23 28.68 28.50 29.57 29.73 29.14
SGIO BUFNO=500 29.51 27.73 27.79 27.88 27.93 28.17
SGIO BUFNO=1000 27.43 28.10 27.54 28.04 28.23 27.87
SGIO BUFNO=5000 28.78 28.37 28.39 28.92 28.74 28.64
SGIO BUFNO=10000 28.68 28.76 28.65 28.67 28.81 28.71[/code:4eozqpif]
Comparison of Data-Set Creation with a Default (4K) Page Size and Varying BUFNO= Option Values on a 32-Bit Platform (Time
Measured in Seconds)
不过我试了下,但运行时间和上面列的时间有比较大的差距 |
|