We demonstrate the scalability of Active Testing by running benchmarks with UPC-Thrille on large clusters with thousands of cores. We evaluate the effectiveness of Active Testing by running our tools on several Java and UPC benchmarks, showing that it can predict and confirm real concurrency bugs with low overhead. There are not any significant differences in the cases of the UTS benchmark. We present extensions to UPC-Thrille that support hybrid memory models as well. We explain in detail the design decisions and optimizations that were necessary to scale Active Testing to thousands of cores. In this section, we rst discuss the design requirements for queues. ited to: Unified Parallel C (UPC) 2, Titanium 3, Coarray Fortran 4, X10 5, and Chapel 6. timized and enhanced implementations of UTS benchmark. We have also developed UPC-Thrille, an Active Testing framework for Unified Parallel C (UPC) programs written in the Single Program Multiple Data (SPMD) programming model combined with the Partitioned Global Address Space (PGAS) memory model. 5.5 Pseudo-code for the UTS benchmark in CAF 2.0. We have implemented Active Testing for multi-threaded Java programs in the CalFuzzer framework. Husbands et al.
#BERKELEY UPC BENCHMARK UTS SOFTWARE#
The BeBOP group is broadly interested in understanding software performance tuning issues, and the interaction or implications for hardware design. It works in two phases: in the predictive analysis phase, a program is executed and monitored for potential concurrency bugs and in the testing phase, Active Testing re-executes the program while controlling the thread schedules in an attempt to reproduce the bug predicted in the first phase. Berkeley Benchmarking and Optimization Home Page. We present Active Testing, a methodology to effectively find concurrency bugs that scales to large distributed memory parallel systems.Īctive Testing combines the coverage and predictive power of program analysis with the familiarity of testing. Hargrove: 'Re: upc-benchmark' Hi, here is my answer to your question. Hargrove: 'Re: upc-benchmark' Reply: Paul H. Performance is comparable to MPI on the NAS Parallel Benchmarks (within a few percent) Several optimization opportunities exist UPC at LBNL/U.C. These bugs are often non-deterministic and hard to find without sophisticated tools. In reply to: luxingjing: 'upc-benchmark' Next in thread: Paul H.
Common classes of concurrency bugs include data races, deadlocks, and atomicity violations. The Berkeley UPC Project Kathy Yelick Christian Bell, Dan Bonachea, Wei Chen, Jason Duell, Paul Hargrove, Parry Husbands, Costin Iancu, Wei Tu, Mike Welcome.
Parallel and concurrent software sometimes exhibit incorrect behavior because of unintended interference between different threads of execution.
0 kommentar(er)
