12733

LightPlay: Efficient Replay with GPUs

Min Feng, Farzad Khorasani, Rajiv Gupta, Laxmi N. Bhuyan
NEC Laboratories America
27th International Workshop on Languages and Compilers for Parallel Computing, 2014

@article{feng2014lightplay,

   title={LightPlay: Efficient Replay with GPUs},

   author={Feng, Min and Khorasani, Farzad and Gupta, Rajiv and Bhuyan, Laxmi N},

   year={2014}

}

Download Download (PDF)   View View   Source Source   

799

views

Previous deterministic replay systems reduce the runtime overhead by either relying on hardware support or by relaxing the determinism requirements for replay. We propose LightPlay that fulfills stricter determinism requirements with low overhead without requiring hardware or OS support. LightPlay guarantees that the memory state after each instruction instance in a replay run is the same as in original run. It reduces logging overhead using a lightweight thread local technique that avoids synchronization between threads during the recording run. GPUs are used to efficiently identify the memory ordering constraints that produce the same memory states before the replay run. LightPlay incurs low space overhead for logging as it only stores the part of log where data races occur. During the logging run LightPlay is 20x-100x faster than logging the total order and requires only 1% space overhead.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2017 hgpu.org

All rights belong to the respective authors

Contact us: