We present a novel approach to dynamic datarace detection for multithreaded object-oriented programs. Past techniques for onthe -fly datarace detection either sacrificed precision for performance, leading to many false positive datarace reports, or maintained precision but incurred significant overheads in the range of 3# to 30#. In contrast, our approach results in very few false positives and runtime overhead in the 13% to 42% range, making it both efficient and precise. This performance improvement is the result of a unique combination of complementary static and dynamic optimization techniques.

We present a new method for dynamically detecting potential data races in multithreaded programs. Our method improves on the state of the art in accuracy, in usability, and in overhead. We improve accuracy by combining two previously known race detection techniques — lockset-based detection and happens-before-based detection — to obtain fewer false positives than lockset-based detection alone. We enhance usability by reporting more information about detected races than any previous dynamic detector. We reduce overhead compared to previous detectors — particularly for large applications such as Web application servers — by not relying on happens-before detection alone, by introducing a new optimization to discard redundant information, and by using a “two phase” approach to identify error-prone program points and then focus instrumentation on those points. We justify our claims by presenting the results of applying our tool to a range of Java programs, including the widely-used Web application servers Resin and Apache Tomcat. Our paper also presents a formalization of lockset-based and happens-before-based approaches in a common framework, allowing us to prove a “folk theorem” that happens-before detection reports fewer false positives than lockset-based detection (but can report more false negatives), and to prove that two key optimizations are correct.

Data race detection is essential for debugging multithreaded programs and assuring their correctness. Nevertheless, there is no single universal technique capable of handling the task efficiently, since the data race detection problem is computationally hard in the general case. Thus, to approximate the possible races in a program, all currently available tools take different “short-cuts”, such as using strong assumptions on the program structure or applying various heuristics. When applied to some general case program, however, they usually result in excessive false alarms or in a large number of undetected races. Another major drawback of many currently available tools is that they are restricted, for perfor-mance reasons, to detection units of fixed size. Thus, they all suffer from the same problem—choosing a small unit might result in missing some of the data races, while choosing a large one might lead to false detection. In this work we present a novel testing tool, called MULTIRACE, which combines improved versions of DJIT and LOCKSET—two very powerful on-the-fly algorithms for dynamic detection of apparent data races. Both extended algorithms detect races in multithreaded programs that may execute on weak consistency systems, and may use two-way as well as global synchronization primitives. By employing novel technologies, MULTIRACE adjusts its detection to the native granularity of objects and variables in the program under examination. In order to monitor all accesses to each of the shared locations, MULTIRACE instruments the C++ source code of the program. It lets the user fine-tune the detection process, but otherwise is completely automatic and transparent. This paper describes the algorithms employed in MULTIRACE, as well as its implementation details. The paper also proposes some alternatives to and optimizations of MULTIRACE. It shows that the overheads imposed by MULTIRACE are often much smaller (orders of magnitude) than those obtained by other existing dynamic techniques.

Distributed Shared-Memory (DSM)computers, which partition physical memory among a collection of workstationlike computing nodes, are now a common way to implement parallel machines. Recently, there has been much interest in DSM machines that use software, instead of hardware, to implement coherence protocols to manage data replication and cache coherence. Software offers many advantages, not the least of which is the possibility of adding significant functionality — such as race detection — to a protocol. This paper describes a new, transparent, protocol-based technique for automatically detecting data races on-the-fly. An implementation of this approach in a DSM system running on a Thinking Machines CM-5 found data races in two of a set of five shared-memory benchmarks. Monitored applications had slowdowns ranging from 0–3 on 32 nodes. 1

Global address space languages like UPC exhibit high performance and portability on a broad class of shared and distributed memory parallel architectures. The most scalable applications use bulk memory copies rather than individual reads and writes to the shared space, but finergrained sharing can be useful for scenarios such as dynamic load balancing, event signaling, and distributed hash tables. In this paper we present three optimization techniques for global address space programs with fine-grained communication: redundancy elimination, use of split-phase communication, and communication coalescing. Parallel UPC programs are analyzed using static single assignment form and a dataflow graph, which are extended to handle the various shared and private pointer types that are available in UPC. The optimizations also take advantage of UPC’s relaxed memory consistency model, which reduces the need for cross thread analysis. We demonstrate the effectiveness of the analysis and optimizations using several benchmarks, which were chosen to reflect the kinds of finegrained, communication-intensive phases that exist in some larger applications. The optimizations show speedups of up to 70 % on three parallel systems, which represent three different types of cluster network technologies. 1

Abstract. Multithreaded programs are prone to errors caused by unintended interference between concurrent threads. This paper focuses on verifying that deterministically-parallel code is free of such thread interference errors. Deterministically-parallel code may create and use new threads, via fork and join, and coordinate their behavior with synchronization primitives, such as barriers and semaphores. Such code does not satisfy the traditional non-interference property of atomicity (or serializability), however, and so existing atomicity tools are inadequate for checking deterministically-parallel code. We introduce a new non-interference specification for deterministically-parallel code, and we present a dynamic analysis to enforce it. We also describe SingleTrack, a prototype implementation of this analysis. SingleTrack’s performance is competitive with prior atomicity checkers, but it produces many fewer spurious warnings because it enforces a more general noninterference property that is applicable to more software. 1

Previously we presented atomic sets, memory locations that share some consistency property, and units of work, code fragments that preserve consistency of atomic sets on which they are declared. We also proposed atomic-set serializability as a correctness criterion for concurrent programs, stating that units of work must be serializable for each atomic set. We showed that a set of problematic data access patterns characterize executions that are not atomic-set serializable. Our criterion subsumes data races (single-location atomic sets) and serializability (all locations in one set). In this paper, we present a dynamic analysis for detecting violations of atomic-set serializability. The analysis can be implemented efficiently, and does not depend on any specific synchronization mechanism. We implemented the analysis and evaluated it on a suite of real programs and benchmarks. We found a number of known errors as well as several problems not previously reported.

In this State of the art Report (SotA), we will give an introduction to work presented in the area of debugging large software systems with modern hardware architectures. We will discuss techniques used for single- multi- and distributed systems. In addition we will provide pointers to work by large players in the field, and major conferences of importance.

Abstract. We study the problem of monitoring concurrent program runs for atomicity violations. Unearthing fundamental results behind scheduling algorithms in database control, we build space-efficient monitoring algorithms for checking atomicity that use space polynomial in the number of active threads and entities, and independent of the length of run monitored. Second, by interpreting the monitoring algorithm as a finite automaton, we solve the model checking problem for atomicity of finite-state concurrent models. This remedies incorrect proofs published in the literature. Finally, we exhibit experimental evidence that our atomicity monitoring algorithm gives substantial time and space benefits on benchmark applications. 1

This paper presents a novel analysis framework and algorithm for statically identifying dataraces in multithreaded object-oriented programs. The framework shows how datarace analysis can be formulated as a conjunction of interthread control flow analysis and points-to analysis of thread objects, synchronization objects and access objects. This formulation can be used to identify a spectrum of dataraces depending on the precision of points-to and control flow information received as input. The framework can be used for datarace analysis of programs written in any multithreaded object-oriented language that supports creation of thread objects, monitor-like synchronization of threads via object-based locking, and global memory accesses via static and instance fields.