This creates an index file with the '.idx' appended to the name of the input file:

#! perl -slw
use strict;

open INDEX, '>:raw', "$ARGV[ 0 ].idx" or die $!;
syswrite INDEX, pack( 'N', 0 ), 4;
syswrite INDEX, pack( 'N', tell *ARGV ), 4 while <>;
close INDEX;
[download]

And this loads the appropriate index file for its input argument and the reads 100 records at random:

#! perl -slw
use strict;
use Time::HiRes qw[ time ];

our $N //= 100;

open INDEX, '<:raw', "$ARGV[ 0 ].idx" or die $!;
my $len = -s( INDEX );
sysread INDEX, my( $idx ), $len;
close INDEX;

my $start = time;
open DAT, '<', $ARGV[ 0 ] or die $!;
for( 1 .. $N ) {
    my $toRead = int rand( length( $idx ) / 4 );
    my $offset = unpack 'N', substr $idx, $toRead * 4, 4;
    seek DAT, $offset, 0;
    my $line = <DAT>;
#    print $line;
}
close DAT;

printf "Ave. %.6f seconds/record\n", ( time() -$start ) / $N;
[download]

And here is a console log with timings of indexing a 1gb file containing 16 million records and then reading a 100 records at random via that index:

[23:03:42.25] c:\test>indexFile 1GB.csv

[23:05:08.24] c:\test>readIndexedFile 1GB.csv
Ave. 0.003699 seconds/record

[23:05:40.38] c:\test>readIndexedFile 1GB.csv
Ave. 0.003991 seconds/record
[download]