RACES 2012 - WORKSHOP ON RELAXING SYNCHRONIZATION FOR MULTICORE AND MANYCORE SCALABILITY

Massively-parallel systems are coming: core counts keep rising -- whether conventional cores as in multicore and manycore systems, or specialized cores as in GPUs. Conventional wisdom has been to utilize this parallelism by reducing synchronization to the minimum required to preserve determinism -- in particular, by eliminating data races. However, Amdahl's law implies that on highly-parallel systems even a small amount of synchronization introduces serialization that limits scaling. Thus, the conventional wisdom is doomed to fail as it hits "the CAS ceiling". We are forced to confront the trade-off between synchronization and the ability of an implementation to scale performance with the number of processors: synchronization inherently limits parallelism.
A new school of thought is arising: one that accepts and even embraces nondeterminism (including data races), and is in return able to dramatically reduce synchronization, or even eliminate it completely. However, this approach requires that we leave the realm of the certain and enter the realm of the merely probable. How can we cast aside the certainty of truth, the security of correctness, the logic of a proof, and adopt a new way of thinking, where answers are good enough but not certain, and where many processors work together in parallel without quite knowing the states that the others are in? We may need some amount of synchronization, but how much? Or better yet, how little? What mental tools and linguistic devices can we give programmers to help them adapt to this challenge? This workshop focuses on these questions and related ones--harnessing parallelism by limiting synchronization, even to the point where programs will compute inconsistent or approximate rather than exact answers.
This workshop aims to bring together researchers who, in the quest for scalability, have been exploring the limits of how much synchronization can be avoided. We will consider the workshop successful if the attendees come away with new insights into fundamental principles, and new ideas for algorithms, data structures, programming languages, and mental models. The goal of this workshop is both to influence current programming practice and to initiate the coalescence of a new research community giving rise to a new subfield within the general area of concurrent and parallel programming. Results generated by the workshop will be made persistent via a workshop website and possibly via the ACM Digital Library.
Submission date: Monday, August 6, 2012
Urlhttp://soft.vub.ac.be/races/