Stride: Search-based deterministic replay in polynomial time via bounded linkage

Abstract

Abstract-Deterministic replay remains as one of the most effective ways to comprehend concurrent bugs. Existing approaches either maintain the exact shared read-write linkages with a large runtime overhead or use exponential off-line algorithms to search for a feasible interleaved execution. In this paper, we propose Stride, a hybrid solution that records the bounded shared memory access linkages at runtime and infers an equivalent interleaving in polynomial time, under the sequential consistency assumption. The recording scheme eliminates the need for synchronizing the shared read operations, which results in a significant overhead reduction. Comparing to the previous state-of-the-art approach of deterministic replay, Stride reduces, on average, 2.5 times of runtime overhead and produces, on average, 3.88 times smaller logs.

Debugging non-deadlock concurrency bugs

“…To do this, the technique will use the bug report of Step 3. In particular, the technique will utilize information about the bugs, such as clusters of memory-access patterns and failing execution traces recorded using a record and replay technique (e.g., [23]). The approach will be to explore the interleavings surrounding the bug fix locations and check whether a new interleaving can trigger the patterns that trigger the same bug patterns.…”

Towards effective and efficient search-based deterministic replay

“…To mitigate the runtime slowdown, alternative approaches have been recently attempted. In particular, search-based solutions avoid recording all memory non-deterministic events and rely on a post-recording inference phase to compute, starting from a partial log, a feasible fault-inducing trace [15,12,1,10,8,16]. The drawback is that replay determinism is not guaranteed, mainly because the lack of information on the exact interleaving of threads shared accesses typically causes an exponential growth of the search space during replay [11].…”

“…On the other hand, PRES [12], CoopREP [10], and Stride [16] incur smaller inference times, at a cost of recording, respectively, the global execution order of every basic block containing shared variables (PRES-BB), the local order of access to a subset of shared program elements, and a relaxed order between shared memory reads and writes.…”

CARE: cache guided deterministic replay for concurrent Java programs

“…However, such information is still insufficient for replaying a concurrent execution on a multiprocessor machine, due to non-deterministic outcomes of parallel shared memory accesses. According to the methodology of tackling shared memory non-determinism, existing multiprocessor deterministic replay work can be roughly classified into two categories: search-based [3,19,29,33,36] and order-based [9,13,18,35,37]. Search-based techniques emphasize reduced record cost, thereby their recorded information is typically incomplete for a faithful replay.…”

“…However, their record cost is still unsatisfactory. For example, Stride [37] generates over 30MB/s (2.5TB/day) of log in its evaluation. For programs that contain mass of shared memory accesses, order-based replay technique can easily generate hundreds of megabytes of log per second, and nobody in practice can afford such performance degradation that may last unpredictably long.…”