You have several options, then.

You can use two separate programs. With those, you can use an anonymous pipe on the command line. You could also use a named pipe or a Unix socket. You could even use Berkeley sockets (the canonical implementation of a TCP/IP API).

You can use threads or fork a new process and use a producer/consumer model. You can then use a pipe, sockets, or shared memory. In the case of threads you also have other options.

You could have one program open another via a piped open call or use IPC::Open2 (or IPC::Open3, if needed).

An alternative is to process your input in chunks and to process your output in chunks, then put your subroutines in a loop. Here's one example of that: while ( <> ) { output( input( $_ ) ) } It really matters what type of data you're dealing with whether or not this would work.