Nested, but Separate: Isolating Unrelated Critical Sections in Real-Time Nested Locking

James Robb, Björn B. Brandenburg

Prior work has produced multiprocessor real-time locking protocols that ensure asymptotically optimal bounds on priority inversion, that support fine-grained nesting of critical sections, or that are independence-preserving under clustered scheduling. However, while several protocols manage to come with two out of these three desirable features, no protocol to date accomplishes all three. Motivated by this gap in capabilities, this paper introduces the Group Independence-Preserving Protocol (GIPP), the first protocol to support fine-grained nested locking, guarantee a notion of independence preservation for fine-grained nested locking, and ensure asymptotically optimal priority-inversion bounds. As a stepping stone, this paper further presents the Clustered k-Exclusion Independence-Preserving Protocol (CKIP), the first asymptotically optimal independence-preserving k-exclusion lock for clustered scheduling. The GIPP and the CKIP rely on allocation inheritance (a.k.a. migratory priority inheritance) as a key mechanism to accomplish independence preservation.

The paper will be presented in the session
Locks, neural networks and resilience – Wednesday, July 8, 15:50 – 16:50 (CET)

Please note, all rights on the videos remain with the authors

Comments are closed.