We propose a SAT-based bounded verification technique, called TCBMC, for threaded C programs. Our work is based on CBMC, which models sequential C programs in which the number of executions for each loop and the depth of recursion are bounded. The novelty of our approach is in bounding the number of context switches allowed among threads. Thus, we obtain an efficient modeling that can be sent to a SAT solver for property checking. We also suggest a novel technique for modeling mutexes and Pthread conditions in concurrent programs. Using this bounded technique, we can detect bugs that invalidate safety properties. These include races and deadlocks, the detection for which is crucial for concurrent programs.

Abstract. This paper presents a method for taking advantage of the efficiency of symbolic model checking using disjunctive partitions, while keeping the number and the size of the partitions small. We define a restricted form of a Kripke structure, called an or-structure, for which it is possible to generate small disjunctive partitions. By changing the image and pre-image procedures, we keep even smaller partial disjunctive partitions in memory. In addition, we show how to translate a (software) program to an or-structure, in order to enable efficient symbolic model checking of the program using its disjunctive partitions. We build one disjunctive partition for each state variable in the model directly from the conjunctive partition of the same variable and independently of all other partitions. This method can be integrated easily into existing model checkers, without changing their input language, and while still taking advantage of reduction algorithms which prefer conjunctive partitions.

No abstract

Abstract-This paper explores the computation and communication overlap capabilities enabled by the new CORE-Direct hardware capabilities introduced in the InfiniBand (IB) Host Channel Adapter (HCA) ConnectX-2. These capabilities enable the progression and completion of data-dependent communications sequences to progress and complete at the network level without any Central Processing Unit (CPU) involvement. We use the latency dominated nonblocking barrier algorithm in this study, and find that at 64 process count, a contiguous time slot of about 80 percent of the nonblocking barrier time is available for computation. This time slot increases as the number of processes participating increases. In contrast, CPU based implementations provide a time slot of up to 30 percent of the nonblocking barrier time. This bodes well for the scalability of simulations employing offloaded collective operations. These capabilities can be used to reduce the effects of system noise, and when using nonblocking collective operations have the potential to hide the effects of application load imbalance.

No abstract