If the input data file is really big, it can be worth using named_pipes so you don't have to write copies of the data to the file system.

You'd need a process to seek to the correct point in the file and then write blocks of data to the named pipe, and then the java program can open the named pipe as a regular file and just read from it. The details vary depending which operating system you're on, and it might be too much effort for your particular problem but it can be a useful technique.

IIRC named pipes are a lot easier to work with on Linux/Unix than on Windows, but I haven't used them on Windows for some years - so things might have got better since then.