Exploring hybrid cache locking to balance performance and time predictability

This paper explores hybrid cache locking, a method that allows a fraction of cache lines to be locked; while the remaining cache lines remain unlocked be locked. With hybrid cache locking, performance and time predictability can be adjusted and tuned as compared to the two extremes, i.e., the regular cache without locking and the traditional cache locking that can potentially lock all cache lines. Our evaluation indicates that hybrid cache locking can often make better tradeoffs between performance and time predictability, or provide more options as compared to either pure caching or pure locking. In some cases we observe that better performance can be achieved by locking only part of the cache.