Browse Prior Art Database

Detection of Outliers for Predictive Failure Analysis of DASD

IP.com Disclosure Number: IPCOM000106484D
Original Publication Date: 1993-Nov-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 2 page(s) / 43K

Publishing Venue

IBM

Related People

Smith, GJ: AUTHOR

Abstract

This invention provides an algorithmic means of detecting abnormal head/disk interfaces in a file that are likely to fail by using an outlier detection scheme.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Detection of Outliers for Predictive Failure Analysis of DASD

      This invention provides an algorithmic means of detecting
abnormal head/disk interfaces in a file that are likely to fail by
using an outlier detection scheme.

      The problem to be solved is how to detect abnormal head/disk
spacing changes in the presence of global spacing changes.  The
solution is to deal with the problem by considering the abnormal
measurements as outliers.  Measurements of head flying height change
can be inferred from changes in the frequency content of the readback
signal from a head using the Wallace equation.  It is common
knowledge that in files with multiple head/disk interfaces, a single
head typically crashes before the others.

      Using the above information, a simple outlier test can be used
to detect a single head that is changing flying height differently
than the others.  To use this method, the nominal value for the
flying height measurement is saved when a direct access storage
device (DASD) is manufactured.  Periodically, new measurements are
made and compared to the reference values.  A statistical test, as
described below, can be used to identify a rogue head.  If an outlier
is identified, a special return code is given to the host system and
the pending crash can be avoided by replacing the DASD.

      The sequence of operations for calculating the statistic, PHI,
is shown next.  A threshold is set for PHI such that if exceeded, the
host system is notified of a pending crash.

1.  Compute the average difference between current flyheight values
  ...