Two-Phase Update for Scalable Concurrent Data Structures

Parallel shared-memory software must control concurrent access to shared resources. This is typically accomplished via locking designs that result in low concurrency or that are highly complex. This paper presents two-phase update, which is an alternative method of updating concurrent data structures, and demonstrates that it is both simple and highly scalable in restricted but commonly occurring situations. Example uses are taken from Sequent’s (now IBM’s) DYNIX/ptx operating system and from the Tornado and K42 research projects. Two-phase update techniques have been used in production by DYNIX/ptx since 1993, and were independently developed by Tornado and K42.