| LDR | | 00000nam u2200205 4500 |
| 001 | | 000000434321 |
| 005 | | 20200226145519 |
| 008 | | 200131s2019 ||||||||||||||||| ||eng d |
| 020 | |
▼a 9781085644884 |
| 035 | |
▼a (MiAaPQ)AAI27528223 |
| 035 | |
▼a (MiAaPQ)NCState_Univ18402036800 |
| 040 | |
▼a MiAaPQ
▼c MiAaPQ
▼d 247004 |
| 082 | 0 |
▼a 621.3 |
| 100 | 1 |
▼a Das, Anwesha. |
| 245 | 10 |
▼a Predicting Location and Time of Anomalies in Large-Scale Computing Systems via Log Mining. |
| 260 | |
▼a [S.l.]:
▼b North Carolina State University.,
▼c 2019. |
| 260 | 1 |
▼a Ann Arbor:
▼b ProQuest Dissertations & Theses,
▼c 2019. |
| 300 | |
▼a 131 p. |
| 500 | |
▼a Source: Dissertations Abstracts International, Volume: 81-03, Section: B. |
| 500 | |
▼a Advisor: Becchi, Michela |
| 502 | 1 |
▼a Thesis (Ph.D.)--North Carolina State University, 2019. |
| 506 | |
▼a This item must not be sold to any third party vendors. |
| 520 | |
▼a Today's large-scale supercomputers encounter faults on a daily basis. Exascale systems are likely to experience even higher fault rates due to increased component count and density. Predicting which node will fail and how soon remains a problem for HPC resilience that needs to be solved to pave the way to exploiting proactive remedies before jobs fail. Triggering resilience-mitigating techniques is still difficult due to the absence ofwell defined failure indicators.Not only for increasing scalability up to exascale systems but even for contemporary supercomputer architectures does it require substantial efforts to distill anomalous events from noisy raw logs. System logs consist of unstructured text that obscures essential system health information contained within. In this context, efficient failure prediction via log mining can enable proactive recovery mechanisms to improve reliability.This thesis makes the following contributions. Two novel solutions are proposed to pin-point node failures, which are unprecedented. First, a phrase extraction-style mechanism, called TBP (time-based phrases) demonstrates the feasibility to predict imminent node failures in Cray systems. Second, a deep learning-based framework, called Desh, predicts short lead times to failures via long short-termmemory (LSTM) networks. These open up the door for enhancing prediction lead times for supercomputing systems in general, thereby facilitating efficient usage of both computing capacity and power. Next, an auto-generated inference scheme, called Aarohi, is developed to achieve speed up during online prediction. Aarohi's parsing is designed to be generic, adaptive, and scalable making it suitable for real-time inference. This compiler-based approach provides a fresh perspective for lead time optimization with a significant prediction speedup required for the deployment of proactive fault tolerant solutions in practice. Further, root cause analysis of node failures is explored using an integrated measurement driven approach to better understand how nodes fail. Our empirical observations about environmental influence and the application effect on failures along with feasible lead time enhancements can facilitate better failure handling in production systems. Finally, we discuss our efforts to deploy the developed failure prediction solutions in a realistic setting to be able to circumvent the hurdles if any. The thesis concludes with a discussion of future research directions in the context of proactive fault tolerance and sustained resilience for large-scale computing systems. |
| 590 | |
▼a School code: 0155. |
| 650 | 4 |
▼a Computer science. |
| 650 | 4 |
▼a Electrical engineering. |
| 690 | |
▼a 0544 |
| 690 | |
▼a 0984 |
| 710 | 20 |
▼a North Carolina State University. |
| 773 | 0 |
▼t Dissertations Abstracts International
▼g 81-03B. |
| 773 | |
▼t Dissertation Abstract International |
| 790 | |
▼a 0155 |
| 791 | |
▼a Ph.D. |
| 792 | |
▼a 2019 |
| 793 | |
▼a English |
| 856 | 40 |
▼u http://www.riss.kr/pdu/ddodLink.do?id=T15494118
▼n KERIS
▼z 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
| 980 | |
▼a 202002
▼f 2020 |
| 990 | |
▼a ***1008102 |
| 991 | |
▼a E-BOOK |