This paper makes the case for studying and
exploiting the predictability of lock operations in multithreaded
programs.  We discuss several uses of <i>lock prediction</i>
and identify the two key events we believe are useful to
predict: (1) which thread will be the next one to acquire a given
lock and (2) which lock a given thread will acquire next. We
describe multiple prediction algorithms and show that while their
accuracy varies widely, they are likely accurate enough for many
uses.