Encoding the output as protocol buffers does not help much with reducing the size, compared to the CSV format. To further reduce the disk size required for storage, the data is gzipped prior to writing it into the file. This makes the resulting files around 70% smaller. Gzip is a common and well supported format, support for decoding it exists in almost every programming language. If this is not desired for e.g. direct access to the stored data, this can be toggled with the -comp command-line flag.