Browse Prior Art Database

DASD Array with Variable Length Records

IP.com Disclosure Number: IPCOM000104051D
Original Publication Date: 1993-Mar-01
Included in the Prior Art Database: 2005-Mar-18
Document File: 4 page(s) / 104K

Publishing Venue

IBM

Related People

Ouchi, NK: AUTHOR

Abstract

Most DASD array treat data as fixed length units that span the array in integral stripes. Disclosed is a mechanism for variable length data and the storage management mechanism needed to effectively use the DASD array space.

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

DASD Array with Variable Length Records

      Most DASD array treat data as fixed length units that span the
array in integral stripes.  Disclosed is a mechanism for variable
length data and the storage management mechanism needed to
effectively use the DASD array space.

      Most DASD arrays assume that the data cover the DASD units as
integral stripes of N data blocks and P parity blocks.  With variable
length data, this assumption can only be held if the data are
"padded" to fill the rest of the stripe.  The average capacity loss
is half of the stripe capacity divided by the capacity of the average
number of stripes per data element.  If the stripe size is small,
this may be an acceptable loss.  Disclosed is the use of a truncated
stripe where the data element is divided into stripes and the last
stripe is not padded with blocks to complete the stripe, but rather
the parity block is appended immediately to the data blocks.  The
capacity loss is half of a block divided by the capacity of the
average number of blocks per data element.  An example is illustrated
in Fig. 1 where the data element is divided into six blocks.  These
blocks are spread across four data units and one parity unit.  Blocks
1 thru 4 form a complete stripe and blocks 5 & 6 form a truncated
stripe.  These are distributed five units U1 thru U5.  The data
blocks and parity blocks are distributed across the units in
sequential order.  Because some stripes are truncated, the starting
point of the sequence can change depending on the last free block of
DASD.  Thus, a specific unit cannot be designated as the "parity"
unit.  In fact, as will be illustrated later, the data blocks may not
be distributed in numerical sequence of DASD units.

      Fig. 2.  continues the example where data element A is added
with blocks A1 thru A3 and the parity block PA13 and data element C1
thru C7 and parity blocks PC14 & PC57.  When data element A is moved
or deleted fr...