Question

I have a network daemon which receives messages of a fixed length (44 Bytes) on a TCP socket.

I am trying to determine what the best length in bytes I should be reading with sysread is. I can of course do a loop that does a sysread for 44 bytes, but I’d like to come up with an optimal size.

I can see an advantage to not having say, a megabyte worth of data to do substr on, but I can also see why doing a thousand sysread calls can cause slowness.

Is there a good size recommendation for doing sysreads over the public internet?

Edit:
The script gets a bunch of the 44 byte messages, they are queued up.

Answer 1

The larger the better! sysread will return as soon any bytes are available.

Since you’re never guaranteed to get a whole message and since you’re never guaranteed to not have more than one message, you need to have a loop on the Perl side. Since you already have a loop on the Perl side, so you might as well get as much data as possible from the system in one go to avoid needless system calls.

use constant READ_SIZE => 65*1024;

my $buf = '';
while (1) {
   my $rv = sysread($fh, $buf, READ_SIZE, length($buf));
   die if !defined($rv);
   last if !$rv;

   while ($buf =~ s/^(.{44})//s) {
      my $msg = $1;
      process_msg($msg);
   }
}

Pick a size and monitor $buf‘s size. If it often approaches READ_SIZE, increase READ_SIZE.