Kuda: GPU hızlandırılmış ayrık yarış durumu denetleyici

We propose a novel technique by introducing a coprocessor to runtime verifica- tion, ergo reducing the cost of race detection without any hardware extension to mainstream PC environment. The goal of our approach is to offload the high com- putational overhead of traditional race detection to hundreds of cores available at modern GPUs. Existing runtime verification frameworks have been designed to run on the same processing units as the code being monitored and (i) instrumentation and (ii) analysis costs contribute to the slowdown of the program being monitored. The framework we propose allows us to carry out (ii) on separate, dedicated cores. As a result, the program being monitored experiences slowdown due to bookkeeping of events, bottleneck is not caused by race detection. An orthogonal line of work shows that with some inexpensive hardware support, monitoring costs can be reduced to negligible levels. By parallelizing the offloaded work, our experiments show that they run as fast as the program being monitored, on separate computational resources. As a demonstration of concept, we investigate runtime monitoring for concurrency bugs, in particular, data race detection. We use a few CPU threads and a large number of cores on a GPU to minimize the slowdown of the application on which race detection is being run.