Organizers: CS Graduate Student Association and faculty (contact: Jianhua Ruan).
Time: 1:00-2:15 pm, Tue / Wed (as posted below)
Place: NPB 3.108, CS Conference Room
Schedule for Fall 2017
- Tue, 9/26
- David Holland (Zhang lab)
- Hongyu Liu (Liu lab)
- Wed, 10/11
- Mohammad Mejbah ul Alam (Muzahid lab)
- Tue, 10/24
- Mitra Bokaei Hosseini (Niu lab)
- Xue Qin (Wang X. lab)
- Wed, 11/8
- Sergio Zamarripa (Gibson. lab)
- Lee Boyd (Zhang lab)
- Tue, 11/21
- Rajasekhar Chaganti (Boppana lab)
- Wed, 12/6
- Naiwei Liu (Yu lab)
- Kevin Baldor (Niu lab)
Abstract: Today's Relational DBMS runs on assumptions and hardware from the 1970s. OLTP performance can be enhanced by an order-of-magnitude by refactoring the legacy DBMS software architecture to adapt and take advantage of advances in 21st Century computer hardware architecture.
Reference: Harizopoulos, S., Abadi, D. J., Madden, S., & Stonebraker, M. (2008). OLTP through the looking glass, and what we found there. In Sigmod'08 (p. 981-992). New York, New York, USA: ACM Press.
Abstract: Reproducing errors of multithreaded programs is very challenging due to many intrinsic non-deterministic factors. Existing RnR systems achieve significant progress for the performance overhead, but none targets the in-situ setting, in which the replay occurs within the same process as the recording process. Also, most of them cannot achieve the identical replay, which may prevent the reproduce of some errors. This paper presents iReplayer, which targets to in-situ identically replay errors for multithreaded programs. The novel in-situ replay of iReplayer makes it more likely to reproduce errors, eliminates the need of replaying from the beginning of executions, and allows for directly employing common debugging tools for diagnosis, without extra steps to setup the replaying environment. iReplayer involved substantial engineering effort to improve its performance and support highly identical replay. Currently, iReplayer only incurs around 3% performance overhead and is scalable, which allows it to be always enabled in production environment. iReplayer is a drop-in library that runs entirely inside the user space, and does not require hardware modification, customized OS, or program modifications. iReplayer enables a range of possibilities, and this paper presents three examples: two automatic tools for detecting buffer overflows and use-after-free bugs, and one debugging tool that is integrated with GDB.
Abstract: The cost of substantial performance bene?ts in multithreaded programs is increasing programming complexity. Debugging multithreaded program is hard and prone to error. This talk will present design and implementation of two debugging approaches for multithreaded programs: ACT  and AutoCon. ACT overcomes the dif?culty of reproducing concurrency bugs in multithreaded programs. It implements hardware assisted machine learning for detecting bugs on the ?y. AutoCon is a neural network based solution for detecting performance regression due to cache contention. It implements anomaly based detection using hardware performance counters. This approach reduces the need for domain expertise and exhausted searching in diagnosing performance issues in multithreaded programs.
Reference: M. M. U. Alam and A. Muzahid. Production-run software failure diagnosis via adaptive communication tracking. In Proceedings of the 43rd International Symposium on Computer Architecture, ISCA '16, pages 354-366, Piscataway, NJ, USA, 2016. IEEE Press.
Questions and Comments?
Please send emails to firstname.lastname@example.org.