Language Constructs for Transactional Memory

摘要

Building concurrent shared-memory data structures is a notori-ously difficult problem, and so the widespread move to multi-coreand multi-processor hardware has led to increasing interest in lan-guage constructs that may make concurrent programming easier.One technique that has been studied widely is the use of atomicblocks [3] built over transactional memory (TM) [2]: the program-mer marks a section of code as atomic, and the language imple-mentation speculatively executes it using transactions [1]. Transac-tions can run in parallel so long as they access different data. In this talk I'll introduce some of the challenges that I've seenin building robust implementations of this idea. What are the lan-guage design choices that exist? What language features can beused inside atomic blocks, and where can atomic blocks oc-cur? Which uses of atomic blocks should be considered correct,and which uses should be considered "racy"? What are the likelyimpacts of different design choices on performance? What are theimpacts on flexibility for the language implementer, and what arethe impacts on flexibility to the programmer using these constructs? I'll argue that one way of trying to resolve these questions isto be rigorous about keeping the ideas of atomic blocks andTM separate; in practice they've often been conflated (not leastin languages that I've worked on). I'll argue that, when thinkingabout atomic blocks, we should keep a wide range of possibleimplementations in mind (for example, TM, lock inference, orsimply control over pre-emption). Similarly, when thinking aboutTM, we should recognize that it can be exposed to programmersthrough a wide range of abstractions and language constructs.