Composing concurrency control

Abstract

Concurrency control poses significant challenges when composing computations over multiple data-structures (objects) with different concurrency-control implementations. We formalize the usually desired requirements (serializability, abort-safety, deadlock-safety, and opacity) as well as stronger versions of these properties that enable composition. We show how to compose protocols satisfying these properties so that the resulting combined protocol also satisfies these properties. Our approach generalizes well-known protocols (such as two-phase-locking and two-phase-commit) and leads to new protocols. We apply this theory to show how we can safely compose optimistic and pessimistic concurrency control. For example, we show how we can execute a transaction that accesses two objects, one controlled by an STM and another by locking.

Power‐aware pipelining with automatic concurrency control

“…Ziv et al 44 combined locking with Software TM (STM) in different parts of the program, obtaining better performance than just using a lock-based or an STM-based solution. They found that, by using one of the two choices exclusively throughout the application, execution is often sub-optimal.…”

“…Quite a few authors have previously combined optimistic and pessimistic concurrency control mechanisms in the context of Database Systems. Ziv et al combined locking with Software TM (STM) in different parts of the program, obtaining better performance than just using a lock‐based or an STM‐based solution. They found that, by using one of the two choices exclusively throughout the application, execution is often sub‐optimal.…”

Drinking from both glasses

“…Dice et al build a runtime library that supports adaptive lock elision using hardware transactional memory (HTM) and optimistic software execution [16]. Ziv et al formalize a theory for correctly composing different concurrency control protocols in programs [43].…”