Output details

A Black-box Approach to Understanding Concurrency in DaCapo

<07> Systems research is driven by benchmarks: these “accelerate, retard or misdirect energy and innovation” (Blackburn et al OOPSLA'06). The importance of this work lies in its platform-independent analysis of the extent to which Java benchmarks really use concurrency mechanisms (e.g. accessing shared memory). We find that the DaCapo benchmarks (Blackburn et al), widely used in concurrency research, do not stress the mechanisms they supposedly exercise. Our conclusions guide VM and garbage collection researchers designing and optimising many-core, shared memory systems, by helping them understand the concurrency-related characteristics of these benchmarks. This work arose from our EPSRC Mirror-GC project, EP/H026975/1.