Summary: Needed: Foundations for Transactional Memory
Dept. of Computer Science, Technion
"It's all very well in practice, but it will never work in theory"
(French management saying)
Transactional memory is a leading paradigm for designing concurrent applications for tomor-
row's multi-core architectures. It follows and draws much inspiration from earlier research on
concurrent data structures and concurrency control. Quite remarkably, it has succeeded in break-
ing out of the research community, and is being seriously considered by the industry--both as part
of software solutions and as the basis for novel hardware designs.
But this success comes at a price, as every new research paper is now being judged by its imme-
diate relevance and applicability to current technology, while ignoring the long-term development
of foundations for this important area.
So, while there is a large body of practical work, the theoretical principles are still lacking:
There are no agreed-upon concepts, and even the terminology is muddled; there is great confusion
between specifications, policies and implementations; even when they exist, correctness properties
do not distinguish between safety, liveness and performance.
This lack of foundations hinders communication and interaction, both within the community of
researchers investigating concurrent data structures and in its interactions with other communities,