Institutional Repository
| PEDAL: a dynamic analysis tool for efficient concurrency bug reproduction in big data environment | |
| Hu, Y; Yan, J; Choo, KKR | |
| 2016 | |
| Source | CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
 |
| ISSN | 1386-7857 |
| Volume | 19Issue:1Pages:153-166 |
| English Abstract | Concurrency bugs usually manifest under very rare conditions, and reproducing such bugs can be a challenging task. To reproduce concurrency bugs with a given input, one would have to explore the vast interleaving space, searching for erroneous schedules. The challenges are compounded in a big data environment. This paper explores the topic of concurrency bug reproduction using runtime data. We approach the concurrency testing and bug reproduction problem differently from existing literature, by emphasizing on the preemptable synchronization points. In our approach, a light-weight profiler is implemented to monitor program runs, and collect synchronization points where thread scheduler could intervene and make scheduling decisions. Traces containing important synchronization API calls and shared memory accesses are recorded and analyzed. Based on the preemptable synchronization points, we build a reduced preemption set (RPS) to narrow down the search space for erroneous schedules. We implement an optimized preemption-bounded schedule search algorithm and an RPS directed search algorithm, in order to reproduce concurrency bugs more efficiently. Those schedule exploration algorithms are integrated into our prototype, Profile directed Event driven Dynamic AnaLysis (PEDAL). The runtime data consisting of synchronization points is used as a source of feedback for PEDAL. To demonstrate utility, we evaluate the performance of PEDAL against those of two systematic concurrency testing tools. The findings demonstrate that PEDAL can detect concurrency bugs more quickly with given inputs, and consuming less memory. To prove its scalability in a big data environment, we use PEDAL to analyze several real concurrency bugs in large scale multithread programs, namely: Apache, and MySQL.; Concurrency bugs usually manifest under very rare conditions, and reproducing such bugs can be a challenging task. To reproduce concurrency bugs with a given input, one would have to explore the vast interleaving space, searching for erroneous schedules. The challenges are compounded in a big data environment. This paper explores the topic of concurrency bug reproduction using runtime data. We approach the concurrency testing and bug reproduction problem differently from existing literature, by emphasizing on the preemptable synchronization points. In our approach, a light-weight profiler is implemented to monitor program runs, and collect synchronization points where thread scheduler could intervene and make scheduling decisions. Traces containing important synchronization API calls and shared memory accesses are recorded and analyzed. Based on the preemptable synchronization points, we build a reduced preemption set (RPS) to narrow down the search space for erroneous schedules. We implement an optimized preemption-bounded schedule search algorithm and an RPS directed search algorithm, in order to reproduce concurrency bugs more efficiently. Those schedule exploration algorithms are integrated into our prototype, Profile directed Event driven Dynamic AnaLysis (PEDAL). The runtime data consisting of synchronization points is used as a source of feedback for PEDAL. To demonstrate utility, we evaluate the performance of PEDAL against those of two systematic concurrency testing tools. The findings demonstrate that PEDAL can detect concurrency bugs more quickly with given inputs, and consuming less memory. To prove its scalability in a big data environment, we use PEDAL to analyze several real concurrency bugs in large scale multithread programs, namely: Apache, and MySQL. |
| Indexed Type | SCI |
| Keyword | Concurrency Analysis Profiling Dynamic Analysis Bug Reproduction |
| Department | Dalian Univ Technol, Sch Software, Dalian, Peoples R China. Chinese Acad Sci, Inst Software, Beijing, Peoples R China. Univ S Australia, Adelaide, SA 5001, Australia. Key Lab Ubiquitous Network, Dalian, Peoples R China. Serv Software Liaoning Prov, Dalian, Peoples R China. |
| Language | 英语 |
| WOS ID | WOS:000373179800013 |
| Citation statistics | |
| Content Type | 期刊论文 |
| URI | http://ir.iscas.ac.cn/handle/311060/17343 |
| Collection | 中国科学院软件研究所 |
| Recommended Citation GB/T 7714 | Hu, Y,Yan, J,Choo, KKR. PEDAL: a dynamic analysis tool for efficient concurrency bug reproduction in big data environment[J]. CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS,2016,19(1):153-166. |
| APA | Hu, Y,Yan, J,&Choo, KKR.(2016).PEDAL: a dynamic analysis tool for efficient concurrency bug reproduction in big data environment.CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS,19(1),153-166. |
| MLA | Hu, Y,et al."PEDAL: a dynamic analysis tool for efficient concurrency bug reproduction in big data environment".CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS 19.1(2016):153-166. |
| Files in This Item: | ||||||
| File Name/Size | DocType | Version | Access | License | ||
| art%3A10.1007%2Fs105（1654KB） | 开放获取 | License | Application Full Text | |||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment