swMATH ID: 33051
Software Authors: Jun Nakashima; Kenjiro Taura
Description: MassiveThreads: A Thread Library for High Productivity Languages. An efficient implementation of task parallelism is important for high productivity languages. Specifically, it requires a tasking layer that fulfills following requirements: (i) its performance scales to high core counts, and (ii) it is seamlessly integrated into a runtime system that performs inter-node communication and synchronization. More specifically, it should facilitate interactions between tasks and threads dedicated for inter-node communication. There have been many implementations that satisfy (i), but, to the best of our knowledge, none of such systems satisfy both requirements. To address this issue, we propose a thread library called MassiveThreads. It provides not only lightweight threads and a scalable dynamic load-balancing mechanism among CPU cores, but also Pthread-compatible API and I/O semantics. In MassiveThreads, issuing a blocking I/O call triggers a user-level context switch instead of blocking the underlying OS-level thread. These features simplify interactions between tasks and communication threads by instantiating both of them on top of MassiveThreads.
Homepage: https://link.springer.com/chapter/10.1007/978-3-662-44471-9_10
Related Software:
Cited in: 0 Publications