Browse Prior Art Database

DASD Fast Write from Multiple Sources using Journals

IP.com Disclosure Number: IPCOM000115769D
Original Publication Date: 1995-Jun-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 8 page(s) / 303K

Publishing Venue

IBM

Related People

Ouchi, NK: AUTHOR

Abstract

Disclosed is a mechanism for DASD Fast Write where all write operations are to a common buffer and all written data are captured in journals associated with each path to the buffer. The buffer and journals are on failure independent units. Data that may have been lost due to the loss of a buffer are reconstructed by staging the DASD image and applying the journal changes in correct sequence. Described are the mechanisms to reclaim journal space when a track is de-staged back to DASD, to reconstruct the tracks in the event of loss of a buffer, and to effectively request a de-stage to aid in reclaiming space on a journal that is getting full.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 23% of the total text.

DASD Fast Write from Multiple Sources using Journals

      Disclosed is a mechanism for DASD Fast Write where all write
operations are to a common buffer and all written data are captured
in journals associated with each path to the buffer.  The buffer and
journals are on failure independent units.  Data that may have been
lost due to the loss of a buffer are reconstructed by staging the
DASD image and applying the journal changes in correct sequence.
Described are the mechanisms to reclaim journal space when a track is
de-staged back to DASD, to reconstruct the tracks in the event of
loss of a buffer, and to effectively request a de-stage to aid in
reclaiming space on a journal that is getting full.

      Checkpoints and journals have been used by accountants and
information processing systems to reconstruct data in the event of
loss of the prime copy.  Disclosed is a DASD Fast Write system using
the copy of data on the drive as the checkpoint and a journal of
written data.  A similar mechanism has been disclosed using
Non-Volatile Storage units (NVS).  Described is a mechanism that
permits direct writing of data to a common buffer by multiple
independent data paths each with a journal; mechanisms in the journal
and buffer for rapid journal reclamation and reconstruction of data
when the common buffer is lost; use of failure independent common
buffer and journals that removes the need for NVS.  This mechanism
differs from the DASD Fast Write used in the IBM*3990 where data are
written to a common buffer (cache) by independent paths (channels)
then transferred to an NVS.  The 3990 mechanism is fast compared to
writing to DASD but has a delay in transferring to NVS and can be
limited by the data rate of the NVS when multiple paths are
performing DASD Fast Write.  The disclosed DASD Fast Write permits
all of the paths direct write to the common buffer without delay and
at the aggregate data rate of the paths.  The common buffer is
de-staged to DASD when power is turned off.  Batteries power the DASD
and buffer if the primary power fails so that the buffer can be
dumped to DASD.

      Fig. 1 illustrates a DASD system with a set of data sources,
multiple independent paths with journals, a common buffer, and the
DASD devices.  Each path has a journal of all data written to the
buffer through that path.  Each journal is a FIFO queue of finite
length where space is reclaimed by moving the end point, TAIL, as
journal entries are removed.  The mechanism has random access to the
journal to read and write control fields in the journal.  This is
illustrated in Fig. 2 .  The data in the common buffer is organized
as tracks.  While each track can be accessed by any path, only one
path can read or write it at any given time.  The number of tracks
that can have concurrent writing is equal to the number of paths, one
per path.  When a track is de-staged, written back to DASD, the
journal entries in each of the paths are remove...