Skip to main content

TS

Error diagnosis of cloud application operation using bayesian networks and online optimisation

Authors

Sherry Xu, Liming Zhu, Daniel Sun, An Binh Tran, Ingo Weber, Min Fu and Len Bass

NICTA

UNSW

Abstract

Operations such as upgrade or redeployment are an important cause of system outages. Diagnosing such errors at runtime poses significant challenges. In this paper, we propose an error diagnosis approach using Bayesian Networks. Each node in the network captures the potential (root) causes of operational errors and its probability under different operational contexts. Once an operational error is detected, our diagnosis algorithm chooses a starting node, traverses the Bayesian Network and performs assertion checking associated with each node to confirm the error, retrieve further information and update the belief network. The next node in the network to check is selected through an online optimisation that minimises the overall avail- ability risk considering diagnosis time and fault consequence. Our experiments show that the technique minimises the risk of faults significantly compared to other approaches in most cases. The diagnosis accuracy is high but also depends on the transient nature of a fault.

BibTeX Entry

  @inproceedings{Xu_ZSTWFB_15,
    author           = {Xu, Xiwei (Sherry) and Zhu, Liming and Sun, Wei (Daniel) and Tran, An Binh and Weber, Ingo and Fu,
                        Min and Bass, Len},
    month            = sep,
    year             = {2015},
    keywords         = {diagnosis; cloud; operation; bayesian network; online optimisation},
    title            = {Error Diagnosis of Cloud Application Operation Using Bayesian Networks and Online Optimisation},
    booktitle        = {EDCC2015},
    address          = {Paris, France}
  }

Download

Served by Apache on Linux on seL4.