Deconstructing Transactional Semantics: The Subtleties of Atomicity
Colin Blundell, E. Christopher Lewis, Milo M. K. Martin · 2005
Researchers have recently proposed software and hardware support for transactions as a replacement for the traditional lock-based synchronization most common in multithreaded programs. Transactions allow the programmer to specify a region of the program that should appear to execute atomically, while the hardware and runtime system optimistically execute the transactions concurrently to obtain high performance. The transactional abstraction is thus a promising approach for creating both faster and simpler multithreaded programs. Although transactions have great potential for simplifying multithreaded programming due to their strong atomicity guarantees, this work shows that these same guarantees can have unexpected and potentially serious negative effects on programs that were written assuming weaker synchronization primitives. We make three contributions: (1) we show that a direct translation (statically or dynamically) of lock-based critical sections into transactions can introduce deadlocks into otherwise correct programs, (2) we define an atomicity model for transactions, in which we introduce the terms strong and weak atomicity, and (3) we show that the decision to enforce strong atomicity as opposed to weak atomicity can also result in deadlock. These results invalidate the intuitive idea that transactions are strictly safer than lock-based critical sections and strong atomicity is strictly safer than weak atomicity. We assert that the research community must confront these subtle issues of transactional semantics by exploring the design space and deciding upon the most appropriate semantics for future transactional systems.