GNUPLOT

-не так Часто Задаваемые Вопросы-

обновлено 2010/12/26
Домашняя страница | Введение | Информация | Галерея | Японский | Английский

Analyze and count httpd access_log

Because an access-log file of Web server becomes very large within a short period, this file is renewed at the fixed interval - every week / month, etc. Here we assume the case that the log-file is renewed every month automatically. This is a very common case for many WWW servers.

A default log-file of Apache (access_log) has following formatted lines. 

host.domain - - [01/Jan/2000:01:23:45 +0900] "GET /index.html HTTP/1.1" 200 1548
host.domain - - [01/Jan/2000:01:23:50 +0900] "GET /icons/mail.png HTTP/1.1" 200 229

Now let's count how many accesses are there during 24 hours. What we need is the "date" part (01) in [01/Jan/2000: . Count the number of lines those "date" is the same. This is done for the first day to 31th in a month. 

#!/usr/bin/perl
while(<>){
    if(/\.html/){
        split;
        $day = substr($_[3],1,2);
        $count[$day]++;
    }
}
for($i=1;$i<=$#count;$i++){
    printf("%10d %10d\n",$i,$count[$i]);
}

In the access_log file, any kinds of Web access such as image files are recorded, so you need to exclude any log-lines other than the access to HTML files. You can also count an access to a definite file by changing if line in the Perl program.

Firstly each line is separated into items (delimiter is white-space), then the "date" part is cut off by substr. Substitute this into the variable $day, and increment its counter. We named the Perl script above "webplot.pl".

This is an example of access statistics to some Web Server in January, 2000, processed by "webplot.plt".

         1       172
         2       321
         3       208
         4       279
         5       327
      ....      ....
        25       588
        26      1038
        27       848
        28       772
        29       570
        30       495
        31       548

The following shows a graph drawn by gnuplot, dumb terminal. The letter < in "< webplot.pl access_log" means to read an output of Perl program.

gnuplot> set term dumb
Terminal type set to 'dumb'
Options are 'feed 79 24'
gnuplot> plot "< webplot.pl access_log" with step


 1600 ++--------+---------+---------+---------+---------+---------+--------++
      +         +         +         +      "< webplot.pl access_log" ****** +
 1400 ++                                            ***                    ++
      |                                             * *                     |
      |                                             * *                     |
 1200 ++                                            * *                    ++
      |                                             * *                     |
 1000 ++                                  ***       * *   ***              ++
      |                                   * *       * *   * *               |
      |                                   * *       * *   * ***             |
  800 ++                                  * *       * *   *   ***          ++
      |                                   * * ***   * *   *     *           |
  600 ++                    ***           * *** *   * *   *     *          ++
      |             ***     * *           *     *** * *****     *** *       |
      |             * *     * *******     *       * *             ***       |
  400 ++            * *     *       *** ***       ***                      ++
      |   ***   *** * * *****         * *                                   |
  200 ++  * ***** *** ***             ***                                  ++
      | ***                                                                 |
      +         +         +         +         +         +         +         +
    0 ++--------+---------+---------+---------+---------+---------+--------++
      0         5        10        15        20        25        30        35
up