Abstract

The advent of multi-core processors means that exploiting parallelism is key to increasing the performance of programs. Many researchers have studied the use of atomic blocks as a way to simplify the construction of scalable parallel programs. However, there is a large body of existing lock-based code, and typically it is incorrect to simply replace lock-based critical sections withatomic blocks. Some problems include the need to do IO within critical sections; the use of primitives such as condition variables; and the sometime reliance on underlying lock properties such as fairness or priority inheritance. In this paper we investigate an alternative: a software runtime system that allows threads to speculatively execute lock-based critical sections in parallel. Execution proceeds optimistically, dynamically detecting conflicts between accesses by concurrent threads. However, if there are frequent conflicts, or if there are attempts to perform operations that cannot be done speculatively, then execution can fall back to acquiring a lock. Conversely, implementations ofatomic blocks must typically serialise all operations that cannot be performed speculatively. Our runtime system has been designed with the requirements of systems code in mind: in particular it does not require that programs be written in type-safe languages, nor does it require any form of garbage collection. Furthermore, we never require a thread holding a lock to wait for a thread that has speculatively acquired it. This lets us retain any useful underlying properties of a given lock implementation, e.g. fairness or priority inheritance.

Citations