It makes no sense to read a large file a byte at a time, on a system where you use perl. 3GB is a bit much to hold in memory at once, at least it is for me, but a buffer of a few k to a few tens of k should work really well. Also, there's no need for the two handles... use the same buffer for both.

My idea is to fill the buffer, look for the start character, and if you find it too close to the end of the block, read some more and append it to the buffer. read() even supports this operation out of the box.

Sample code (not tested well):

my $windowsize = 500;
use constant BLOCKLENGTH => 4096;
my $buffer = "";
my $offset = 0;
while(my $r = read FH, $buffer, BLOCKLENGTH) {
    my $i = -1;
    until(($i = index($buffer, "x", ++$i)) < 0) {
        printf "found 'x' at %d+%d\n", $offset, $i;
        if($i+$windowsize > length $buffer) {
            # get rid of what we no longer need, or we might end up wi
+th a buffer holding the whole huge file:
            $offset += $i;
            $buffer = substr $buffer, $i;
            $i = 0;
            # append a new buffer (assuming BLOCKLENGTH >= $windowsize
+):
            $r = read FH, $buffer, BLOCKLENGTH, length $buffer;
            last if $windowsize > length $buffer; #not long enough
        }
        # do something here...
        printf "offset for 'x' is %d, found '%s' at %d\n",
          $offset + $i, substr($buffer, $i + $windowsize - 1, 1),
          $offset+$i+$windowsize-1;
    }
} continue {
    $offset += length $buffer;
}
[download]