Paper¶
Algorithms & Models¶
2014-Microsoft-Adtributor¶
Adtributor1最早系统地提出利用根因分析对广告系统收入指标进行溯因， 其基于一个较强的假设：根因的指标来自于单个指标。
2016-Microsoft-iDice¶
iDice2对Adtributor1中的根因位于单个维度 的假定进行了放宽。在iDice中, 允许根因是多个维度的组合。
2018-Baidu-HotSpot¶
HotSpot3指出多维根因分析的两个难点：单个指标的异常会传播导致该指标在不同层级的异常； 算法搜索空间过大，需要高效的搜索算法。针对这两个难点，论文给出了对应的解决方案：对于第一个异常传播的问题，提出了 一个新的指标ripple effect用于得分计算; 对于第二个问题采用蒙特卡洛搜索树(Monte Carlo Tree Search)和层次剪枝(hierarchical pruning)的方法 来实现更加高效的搜索。
2019-BizSeer-Squeeze¶
Squeeze4提出generalized ripple effect和generalized potential score, 同时可以更好地平衡搜索效率与精度.
2021-CAS-AutoRoot¶
AutoRoot5使用daptive density clustering来提升模型精度, 同时使用一种高效的过滤机制来提升搜索效率.
2022-Huawei-RiskLoc¶
RiskLoc6通过加权的方式定义risk score来挖掘根因指标。
2022-Microsoft-CMMD¶
CMMD7主要由两个部分组成： relationship modeling, 根据历史数据用GNN来构建指标之间的关联关系; root cause localization, 使用遗传算法(genetic algorithm)来高效准确地定位根因.
-
R. Bhagwan et al., “Adtributor: Revenue debugging in advertising systems,” in 11th USENIX symposium on networked systems design and implementation (NSDI 14), 2014, pp. 43–55. ↩↩
-
Q. Lin, J.-G. Lou, H. Zhang, and D. Zhang, “iDice: Problem identification for emerging issues,” in Proceedings of the 38th international conference on software engineering, 2016, pp. 214–224. ↩
-
Y. Sun et al., “Hotspot: Anomaly localization for additive kpis with multi-dimensional attributes,” IEEE Access, vol. 6, pp. 10909–10923, 2018. ↩
-
Z. Li et al., “Generic and robust localization of multi-dimensional root causes,” in 2019 IEEE 30th international symposium on software reliability engineering (ISSRE), IEEE, 2019, pp. 47–57. ↩
-
P. Jing, Y. Han, J. Sun, T. Lin, and Y. Hu, “AutoRoot: A novel fault localization schema of multi-dimensional root causes,” in 2021 IEEE wireless communications and networking conference (WCNC), IEEE, 2021, pp. 1–7. ↩
-
M. Kalander, “RiskLoc: Localization of multi-dimensional root causes by weighted risk,” arXiv preprint arXiv:2205.10004, 2022. ↩
-
S. Yan et al., “CMMD: Cross-metric multi-dimensional root cause analysis,” arXiv preprint arXiv:2203.16280, 2022. ↩