[rrd-users] Performance problem in RRDp.pm

[Bjorn Nordbo]

I hope this is the right place to ask.

I have hacked rrdtool into MRTG for normal operation (ie. for batch
generation of graphs). This seems to work relatively ok now, but it
is not by far as fast as I had hoped.

I use RRDp.pm to talk with rrdtool, and the performance problem is
in this module, more specificly the read function. I have tried to
understand this code, and though I see that it uses non-blocking IO,
its operation is not entirely clear to me.

I anyone could give me a summary of operation for this function, I
would be most happy. And if anyone has suggestions for performance
improvmance, please tell me! :)

This is the results of profiling a large session with SmallProf.pm.
Only RRDp::read() is included. The select itself is not visible, but
the line is marked with stars.

      380 0.000000 0.000000   128:sub read () {
      380 0.439632 0.060000   129:  croak "RRDp::read can only be called after
        0 0.000000 0.000000   130:    unless $Sequence eq 'C';
      380 0.000912 0.020000   131:  $Sequence = 'R';
      380 0.001382 0.000000   132:  my $inmask = 0;
      380 0.004526 0.010000   133:  my $srbuf;
      380 0.001269 0.010000   134:  my $minibuf;
      380 0.001280 0.040000   135:  my $buffer;
      380 0.001952 0.010000   136:  my $nfound;
      380 0.001194 0.030000   137:  my $timeleft;
      380 0.063653 0.030000   138:  my $ERR = 0;
      380 0.002133 0.010000   139:  vec($inmask,fileno(RRDreadHand),1) = 1; #
      380 0.009915 0.010000   140:  while (1) {
      380 0.001219 0.030000   141:    my $rout;
****  380 329.1268 0.060000   142:    $nfound =
      380 0.001554 0.030000   143:    if ($nfound == 0 ) {
        0 0.000000 0.000000   144:      # here, we could do something sensible
        0 0.000000 0.000000   145:      next;
        0 0.000000 0.000000   146:    }
      380 0.003149 0.070000   147:    sysread(RRDreadHand,$srbuf,4096);
      380 0.001077 0.030000   148:    $minibuf .= $srbuf;
      684 0.008893 0.080000   149:    while ($minibuf =~ s|^(.+?)\n||s) {
      684 0.002251 0.050000   150:      my $line = $1;
        0 0.000000 0.000000   151:      # print $line,"\n";
      684 0.005111 0.020000   152:      if ($line =~  m|^ERROR|) {
        0 0.000000 0.000000   153: croak $line;
        0 0.000000 0.000000   154: $ERR = 1;
        0 0.000000 0.000000   155:      }
      684 0.001907 0.020000   156:      elsif ($line =~ m|^OK u:([\d\.]+)
      380 0.001137 0.010000   157:
      380 0.010874 0.040000   158: return $ERR == 1 ? undef : \$buffer;
        0 0.000000 0.000000   159:      } else {
      304 0.002021 0.030000   160: $buffer .= $line. "\n";
        0 0.000000 0.000000   161:      }
        0 0.000000 0.000000   162:    }
        0 0.000000 0.000000   163:  }
        0 0.000000 0.000000   164:}

Thanks in advance!

Bjørn

--
Unsubscribe mailto:rrd-users-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-users-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-users