S.m.a.r.t. threshold optimization method used for disk failure detection

Patent metadata

Field	Value
Publication number	US-2019205193-A1
Application number	US-201816135480-A
Country	US
Kind code	A1
Filing date	Sep 19, 2018
Priority date	Dec 29, 2017
Publication date	Jul 4, 2019
Grant date	—

Abstract

An S.M.A.R.T. threshold optimization method used for disk failure detection includes the steps of: analyzing S.M.A.R.T. attributes based on correlation between S.M.A.R.T. attribute information about plural failed and non-failed disks and failure information and sieving out weakly correlated attributes and/or strongly correlated attributes; and setting threshold intervals, multivariate thresholds and/or native thresholds corresponding to the S.M.A.R.T. attributes based on distribution patterns of the strongly or weakly correlated attributes. As compared to reactive fault tolerance, the disclosed method has no negative effects on reading and writing performance of disks and performance of storage systems as a whole. As compared to the known methods that use native disk S.M.A.R.T. thresholds, the disclosed method significantly improves disk failure detection rate with a low false alarm rate. As compared to disk failure forecast based on machine learning algorithm, the disclosed method has good interpretability and allows easy adjustment of its forecast performance.

First claim

What is claimed is:

1 . A self monitoring analysis and reporting technology (S.M.A.R.T.) threshold optimization method used for disk failure detection, comprising the steps of:
collecting S.M.A.R.T. attributes associated with a plurality of computer disk drives; analyzing the collected S.M.A.R.T. attributes based on correlation between S.M.A.R.T. attribute information about plural failed and non-failed disks and failure information; seperating weakly correlated attributes and strongly correlated attributes; setting within a computer having a disk drive at least one of threshold intervals, multivariate thresholds and/or native thresholds corresponding to the S.M.A.R.T. attributes based on distribution patterns of the strongly correlated attributes and weakly correlated attributes; and changing the disk drive within the computer when one or more of the settings of at least one of threshold intervals, multivariate thresholds and/or native thresholds have been met.

2 . The S.M.A.R.T. threshold optimization method of claim 1 , wherein the method further comprises:
setting at least one of the threshold intervals for one of the strongly correlated attributes based on the distribution patterns of the strongly correlated attributes of the S.M.A.R.T. attributes; and setting at least one of the multivariate thresholds for at least two of the strongly correlated attributes based on the distribution patterns of the strongly correlated attributes of the S.M.A.R.T. attributes.

3 . The S.M.A.R.T. threshold optimization method of claim 2 , wherein the strongly correlated attributes and the weakly correlated attributes are analyzed based on a correlation level between at least one said S.M.A.R.T. attribute and time series and/or frequency.

4 . The S.M.A.R.T. threshold optimization method of claim 3 , wherein the method further comprises the steps of:
based on one-dimensional distribution patterns of one of the strongly correlated attributes of the non-failed and failed disks, setting at least one of the threshold intervals for the strongly correlated attribute; and based on multi-dimensional distribution patterns of at least two of the strongly correlated attributes of the non-failed and failed disks, setting the multivariate thresholds for the strongly correlated attributes.

5 . The S.M.A.R.T. threshold optimization method of claim 4 , wherein the method further comprises:
based on the weakly correlated attributes of the non-failed and failed disks, setting the native thresholds corresponding to the weakly correlated attributes.

6 . The S.M.A.R.T. threshold optimization method of claim 5 , wherein the step of setting the at least one threshold interval for one of the strongly correlated attributes comprises:
collecting the S.M.A.R.T. attribute information of the failed disks and of the non-failed disks, respectively, so as to select positive samples and negative samples associated with the strongly correlated attributes; performing learning training on the positive samples and negative samples based on at least one function so as to build a support vector machine whose false alarm rate is below a false alarm threshold; and selecting at least one distribution range of the negative samples that contains relatively few positive samples as the threshold interval based on a support vector distribution diagram of the strongly correlated attributes and setting early warning.

7 . The S.M.A.R.T. threshold optimization method of claim 5 , wherein the step of setting the multivariate thresholds for at least two of the strongly correlated attributes comprises:
collecting the S.M.A.R.T. attribute information of the failed disks and of the non-failed disks, respectively, so as to select positive samples and negative samples associated with at least two said strongly correlated attributes; performing learning training on the positive samples and negative samples based on at least one function so as to build a support vector machine whose false alarm rate is below a false alarm threshold; plotting a multi-dimensional support vector distribution diagram that sets at least two said strongly correlated attributes based on support vectors of the support vector machine; and selecting a distribution range of at least one said negative sample that contains relatively few positive samples based on the multi-dimensional support vector distribution diagram as the multivariate threshold interval and setting early warning.

8 . The S.M.A.R.T. threshold optimization method of claim 7 , wherein the strongly correlated attributes comprise data that differentiate the S.M.A.R.T. attributes of the non-failed and failed disks and data variations thereof, in which
the data of the S.M.A.R.T. attributes comprise at least one of raw values and/or normalized values, and the data variations of the S.M.A.R.T. attributes comprise at least one of a data variation rate in a certain period of time, a difference between at least two data and/or a ratio between at least two data.

9 . A self monitoring analysis and reporting technology (S.M.A.R.T.) threshold optimization device used for disk failure detection, wherein the device at least comprises a S.M.A.R.T. attribute analyzing unit and a threshold setting unit,
the S.M.A.R.T. attribute analyzing unit receiving SMART attributes associated with a plurality of computer disk drives; the S.M.A.R.T. attribute analyzing unit analyzing the S.M.A.R.T. attributes based on correlation between S.M.A.R.T. attribute information about plural failed and non-failed disks and failure information and identifying weakly correlated attributes and strongly correlated attributes; the threshold setting unit setting in a computer system having a disk drive at least one of threshold intervals, multivariate thresholds and/or native thresholds corresponding to the S.M.A.R.T. attributes based on distribution patterns of the strongly correlated attributes and weakly correlated attributes.

10 . The S.M.A.R.T. threshold optimization device used for disk failure detection of claim 9 , wherein the threshold setting unit comprises a multiple threshold interval setting unit and a multivariate threshold setting unit,
the multiple threshold interval setting unit setting at least one of the threshold intervals for one of the strongly correlated attributes based on the distribution patterns of the strongly correlated attributes of the S.M.A.R.T. attributes; and the multivariate threshold setting unit setting the multivariate thresholds for at least two of the strongly correlated attributes based on the distribution patterns of the strongly correlated attributes of the S.M.A.R.T. attributes.

Assignees

• Univ Huazhong Science Tech

Inventors

• Wu Song
• Xiong Zhuang
• Jin Hai

Classifications

• G06F11/076

  by exceeding a count or rate limit, e.g. word- or bit count limit · CPC title

• G06F11/3457

  Performance evaluation by simulation · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06F18/2411

  based on the proximity to a decision surface, e.g. support vector machines · CPC title

• G06F11/2221

  to test input/output devices or peripheral units · CPC title