Browse Prior Art Database

Scheme for High Data Availability from a Shared-Cache in the Event of Hardware/Software Failures in a Multi-system Shared Disk Environment

IP.com Disclosure Number: IPCOM000114352D
Original Publication Date: 1994-Dec-01
Included in the Prior Art Database: 2005-Mar-28
Document File: 4 page(s) / 177K

Publishing Venue

IBM

Related People

Bhargava, G: AUTHOR [+4]

Abstract

This invention presents a scheme for providing high data-availability from a shared-cache in the presence of system/cast out failures in a multi-system, shared-disk, transaction processing environment.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 27% of the total text.

Scheme for High Data Availability from a Shared-Cache in the Event
of Hardware/Software Failures in a Multi-system Shared Disk Environment

      This invention presents a scheme for providing high
data-availability from a shared-cache in the presence of system/cast
out failures in a multi-system, shared-disk, transaction processing
environment.

      A shared electronic cache can be used for propagating updates
between database management systems (DBMSs) in a multi-system data
sharing complex with shared disks.  Presented in this invention is a
scheme for isolating pages of the shared cache that are impacted by
temporary DBMS or cast out failures.  Schemes for quick recovery of
such pages are also described.

The "timestamp advancing mechanism" can be summarized as follows:
  o  Each DBMS maintains a local timestamp which is earlier than the
      timestamp of any page that has not been externalized.  This
      timestamp is referred to as the sys-timestamp.
  o  When a DBMS updates a page, it writes a log record and
      subsequently writes the page to the shared cache as a changed
      page.  With the write operation it provides its sys-timestamp.
  o  The cast out process computes the min(sys-timestamp) across the
      systems and updates the timestamp in the cache directory entry
of
      the page after it has written the page to disk.
  o  Periodically, by computing the minimum timestamp across a)
      sys-timestamp of all systems before scanning the cache and b)
all
      pages in the cache, it is possible to establish a position in
the
      time-sequenced merged log(s) so that all updates to cached
pages,
      not already reflected in their respective disk versions, can be
      captured through the log records.  Consequently, these pages
can
      be recovered from disk in case the cache fails.  This timestamp
      is called the recover-timestamp.

      In the mainline operations of DBMSs, the cast out process reads
sys-timestamps periodically and computes their minimum.  Even if a
system in the data-sharing complex is down, its sys-timestamp (which
is not progressing) will be taken into account.  This has the effect
of not allowing the recover-timestamp to move forward.  For data
availability reasons, this may be an unacceptable situation; in this
invention, a solution to remedy this situation is presented.

      A similar problem occurs when a page cannot be cast out to disk
because, for example, disk connectivity is temporarily lost.  The
recover timestamp does not move forward because the directory entry
in the cache for such a page still reflects the old value.

      Invention - for isolating failed pages, a page error list is
used.  The purpose of the error list is to track pages for which the
DBMS may disallow read/write access and to track the starting and
ending timestamps used for the recovery of such pages.  T...