Browse Prior Art Database

Fault-Tolerant Control for DASD Arrays

IP.com Disclosure Number: IPCOM000115688D
Original Publication Date: 1995-Jun-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 4 page(s) / 161K

Publishing Venue

IBM

Related People

Ouchi, NK: AUTHOR

Abstract

Disclosed is a DASD array control mechanism with two or more failure independent units each with nonvolatile storage and access to systems and DASD units. It provides fault tolerant access to DASD, protects array created data and provides cross unit recovery on failure. DASD arrays provide a very high level of protection of data loss. The array control must not only protect this data but it must protect the data that it creates and updates. The array control must also provide fault tolerant access to the data.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 39% of the total text.

Fault-Tolerant Control for DASD Arrays

      Disclosed is a DASD array control mechanism with two or more
failure independent units each with nonvolatile storage and access to
systems and DASD units.  It provides fault tolerant access to DASD,
protects array created data and provides cross unit recovery on
failure.  DASD arrays provide a very high level of protection of data
loss.  The array control must not only protect this data but it must
protect the data that it creates and updates.  The array control must
also provide fault tolerant access to the data.

      Advance function DASD arrays create and update array data that
are only in the array.  Parity data or array configuration
information are examples of data that must be protected against
failures.  Data is also updated in the array and such data must be
protected during writing since failures during writing destroys both
the old data and the data being written.  Such data is kept in a
fault tolerant storage, a Nonvolatile Store (NVS).  A set of
semiconductor storage units can be used in conjunction with battery
backup to form a fault tolerant storage necessary for DASD arrays.
Magnetic cores or "Flash EPROMS" are other technologies that may be
used for NVS.  The IBM* 3990 has a semiconductor array that has
battery back-up to prevent data lost due to certain failures, but has
single points of failure that can cause loss of access to data.  The
disclosed mechanism, illustrated in Fig. 1, consists of two or more
failure independent units where each unit has an NVS.  Each unit
connects to all of the DASD units and using systems.  Thus, the
systems have at least two failure independent access paths to the
DASD array.  The failure independent units are connected by two or
more links.  These links are used to recover data from the unit when
it fails and the other unit is still operational.  Similar serial
links
are used by the units for messages to coordinate control of the data.

      To further protect array data during an update operation,
within each unit the NVS is two or more failure independent subunits.
An update operation over writes the old data.  A failure during
update may cause loss of both old and new information.  Note that a
system that keeps the old data on DASD and writes the new data in
another area can be envisioned.  The disclosed mechanism teaches that
the old data can be kept in an NVS until the new data successfully
over writes the old data.  The new data, the old data, and the data
on DASD are a triad during the update operation.  Loss of any one can
be recovered from the other two.  Thus, the NVS has two or more
failure independent subunits.  These subunits must be nonvolatile and
can sustain prolonged power loss.  The requirement for nonvolatility
arises in the case where a record is being updated and power or other
massive failure occurs.  The data on the device is destroyed.  But
the NVS images of the new and old data can...