mahimahi/scripts/mm-throughput-graph

#!/usr/bin/env perl

use warnings;
use strict;
use POSIX;

sub usage
{
  die qq{Usage: $0 MS_PER_BIN [filename]\n};
}

if ( scalar @ARGV < 1 or scalar @ARGV > 2 ) {
  usage;
}

my $MS_PER_BIN = shift;

if ( $MS_PER_BIN !~ m{^\d+$} ) {
  usage;
}

sub ms_to_bin {
  return int( $_[0] / $MS_PER_BIN );
}

sub bin_to_seconds {
  return sprintf q{%.3f}, $_[0] * $MS_PER_BIN / 1000.0;
}

my ( %capacity, %arrivals, %departures );

my $first_timestamp = undef;
my $last_timestamp = undef;
my $base_timestamp = undef;
my $capacity_sum;
my $arrival_sum;
my $departure_sum;
my @delays;
my %signal_delay;

LINE: while ( <> ) {
  chomp;

  if ( m{^# base timestamp: (\d+)} ) {
    if ( defined $base_timestamp ) {
      die "base timestamp multiply defined";
    } else {
      $base_timestamp = $1;
    }
    next LINE;
  } elsif ( m{^#} ) {
    next LINE;
  }

  # parse and validate line
  my ( $timestamp, $event_type, $num_bytes, $delay ) = split /\s+/, $_;

  if ( not defined $num_bytes ) {
    die q{Format: timestamp event_type num_bytes [delay]};
  }

  if ( $timestamp !~ m{^\d+$} ) {
    die qq{Invalid timestamp: $timestamp};
  }

  if ( $num_bytes !~ m{^\d+$} ) {
    die qq{Invalid byte count: $num_bytes};
  }

  if ( not defined $base_timestamp ) {
    die "logfile is missing base timestamp";
  }

  $timestamp -= $base_timestamp; # correct for startup time variation

  if ( not defined $last_timestamp ) {
    $last_timestamp = $first_timestamp = $timestamp;
  }

  $last_timestamp = max( $timestamp, $last_timestamp );

  my $num_bits = $num_bytes * 8;
  my $bin = ms_to_bin( $timestamp );

  # process the event
  if ( $event_type eq q{+} ) {
    $arrivals{ $bin } += $num_bits;
    $arrival_sum += $num_bits;
  } elsif ( $event_type eq q{#} ) {
    $capacity{ $bin } += $num_bits;
    $capacity_sum += $num_bits;
  } elsif ( $event_type eq q{-} ) {
    if ( not defined $delay ) {
      die q{Departure format: timestamp - num_bytes delay};
    }
    $departures{ $bin } += $num_bits;

    if ( $delay < 0 ) {
	die qq{Invalid delay: $delay};
    }

    if ( $timestamp - $delay < 0 ) {
	die qq{Invalid timestamp and delay: ts=$timestamp, delay=$delay};
    }

    push @delays, $delay;
    $departure_sum += $num_bits;

    $signal_delay{ $timestamp - $delay } = min( $delay,
						(defined $signal_delay{ $timestamp - $delay })
						? $signal_delay{ $timestamp - $delay }
						: POSIX::DBL_MAX );
  } else {
    die qq{Unknown event type: $event_type};
  }
}

sub min {
  my $minval = POSIX::DBL_MAX;

  for ( @_ ) {
    if ( $_ < $minval ) {
      $minval = $_;
    }
  }

  return $minval;
}

sub max {
  my $maxval = - POSIX::DBL_MAX;

  for ( @_ ) {
    if ( $_ > $maxval ) {
      $maxval = $_;
    }
  }

  return $maxval;
}

if ( not defined $first_timestamp ) {
    die q{Must have at least one event};
}

# calculate statistics
my $duration = ($last_timestamp - $first_timestamp) / 1000.0;
my $average_capacity = ($capacity_sum / $duration) / 1000000.0;
my $average_ingress = ($arrival_sum / $duration) / 1000000.0;
my $average_throughput = ($departure_sum / $duration) / 1000000.0;

if ( scalar @delays == 0 ) {
  die q{Must have at least one departure event};
}

@delays = sort { $a <=> $b } @delays;

my $pp95 = $delays[ 0.95 * scalar @delays ];

# measure signal delay every millisecond
# = minimum time for a message created at time t to get to receiver
my @signal_delay_samples = sort { $a <=> $b } keys %signal_delay;

for ( my $ts = $signal_delay_samples[ -1 ]; $ts >= $signal_delay_samples[ 0 ]; $ts-- ) {
  if ( not defined $signal_delay{ $ts } ) {
    $signal_delay{ $ts } = $signal_delay{ $ts + 1 } + 1;
  }
}

my @signal_delays = sort { $a <=> $b } values %signal_delay;
my $pp95s = $signal_delays[ 0.95 * scalar @signal_delays ];

printf STDERR qq{Average capacity: %.2f Mbits/s\n}, $average_capacity;
printf STDERR qq{Average throughput: %.2f Mbits/s (%.1f%% utilization)\n}, $average_throughput, 100.0 * $average_throughput / $average_capacity;
printf STDERR qq{95th percentile per-packet queueing delay: %.0f ms\n}, $pp95;
printf STDERR qq{95th percentile signal delay: %.0f ms\n}, $pp95s;

# make graph
my $earliest_bin = min( keys %arrivals, keys %capacity, keys %departures );
my $latest_bin = max( keys %arrivals, keys %capacity, keys %departures );

if ( $earliest_bin == $latest_bin ) {
  die q{MS_PER_BIN is too large for length of trace};
}

my $current_buffer_occupancy = 0;

sub default {
  return defined $_[ 0 ] ? $_[ 0 ] : 0;
}

open GNUPLOT, q{| gnuplot} or die;

print GNUPLOT <<END;
set xlabel "time (s)"
set ylabel "throughput (Mbits/s)"
set key center outside top horizontal
set style fill solid 0.2 noborder
set terminal svg size 1024,560 fixed  fname 'Arial' rounded solid mouse standalone name "Throughput"
set output "/dev/stdout"
END

printf GNUPLOT qq{plot [%f:%f] "-" using 1:2 title "Capacity (mean %.2f Mbits/s)" with filledcurves above x1 lw 0.5, "-" using 1:3 with lines lc rgb "#0020a0" lw 4 title "Traffic ingress (mean %.2f Mbits/s)", "-" using 1:4 with lines lc rgb "#ff6040" lw 2 title "Traffic egress (mean %.2f Mbits/s)"\n},
  $first_timestamp / 1000.0, $last_timestamp / 1000.0, $average_capacity, $average_ingress, $average_throughput;

my $output;

for ( my $bin = $earliest_bin; $bin <= $latest_bin; $bin++ ) {
    my $t = bin_to_seconds( $bin );
    my ( $cap, $arr, $dep ) = map { (default $_) / ($MS_PER_BIN / 1000.0) / 1000000.0 } ( $capacity{ $bin }, $arrivals{ $bin }, $departures{ $bin } );

    $current_buffer_occupancy += default $arrivals{ $bin };
    $current_buffer_occupancy -= default $departures{ $bin };

    $output .= qq{$t $cap $arr $dep $current_buffer_occupancy\n};
}

print GNUPLOT $output;
print GNUPLOT qq{\ne\n};
print GNUPLOT $output;
print GNUPLOT qq{\ne\n};
print GNUPLOT $output;

close GNUPLOT or die qq{$!};