The suggestion from the library author is to not use it in a multithreaded fashion, even if it's possible. The right way is to instantiate a ring per core or thread, since they are so cheap