Section: New Results

Correct and efficient runtime systems

Participants : Nhat Minh Lê, Robin Morisset, Adrien Guatto, Antoniu Pop, Francesco Zappa Nardelli, Albert Cohen.

User-space scheduling and concurrent first-in first-out queues are two essential building blocks of parallel programming runtimes. They are, however, rarely used together since typical schedulers are oblivious to the ordering constraints introduced by buffered communication.

Chase and Lev's concurrent deque is a key data structure in shared-memory parallel programming and plays an essential role in work-stealing schedulers. We provided the first correctness proof of an optimized implementation of Chase and Lev's deque on top of the POWER and ARM architectures: these provide very relaxed memory models, which we exploit to improve performance but considerably complicate the reasoning. We also studied an optimized x86 and a portable C11 implementation, conducting systematic experiments to evaluate the impact of memory barrier optimizations. Our results demonstrate the benefits of hand tuning the deque code when running on top of relaxed memory models.

Based on this early success, we started working on a more global solution using a new lock-free algorithm for stalling and waking-up tasks in a user-space scheduler according to changes in the state of the corresponding queues. The algorithm is portable and correct, since it is written and proven against the C11 memory model. We showed through experiments that it can serve as a keystone to efficient parallel runtime systems.

These efforts underline the parallelizing compilation research for $n$-synchronous languages, and the scalable parallel execution of OpenStream.