On 02/13/2012 02:43 PM, Paul Gilmartin wrote:
On Mon, 13 Feb 2012 11:38:44 -0600, Joel C. Ewing wrote:
It should be possible to just specify data set limits in terms of
data bytes expected or records/average-record-length expected without
regard for tracks, cylinders, extents, or volumes. ...
And the user interface should be simplified. I should be able to
code SPACE=(1,54000000000) and let SDB infer an average
block size and allocation spare me the algebra of factoring the
total space into halfword chunks. This might require a new
alternative TU: a 64-bit (for future growth) extent size in bytes.
(I'd prefer, for legibility, SPACE=(1,54.000.000.000) with European
thousands separators.)
-- gil
Not sure about your reference to half-word chunks. Although there are
16Mi limits on max numerical value for primary-qty parameter,
"SPACE=(1,54000),AVGREC=M"
is perfectly legitimate for allocating 54,000 MiB (only about 5% high if
you really needed exactly 54,000 MB) for a sequential data set, provided
you have a DATACLAS that also specifies EXTENDED, Storage Constraint
Relief, and a high enough volume count. Allocation will spread the
dataset over as many volumes as necessary and up to as large a number
(127?) of extents per volume as necessary to allocate the requested
total space. Although the first value in SPACE is described as average
record length, it is really only used as a multiplier for primary-qty or
secondary-qty and AVGREC for calculating total data bytes needed, under
the assumption the actual BLKSIZE will give efficient track utilization.
The "gotcha" used to be that if you grossly over-requested space, got
space dispersed over umpteen volumes, only used a little of the space,
that "RLSE" would then only release the unused space on the last volume
actually written and leave all the unneeded, unused space on subsequent
volumes allocated until the data set was deleted. One would hope that
issue would eventually be resolved and the concept would be extended to
data set types other than DSORG=PS.
--
Joel C. Ewing, Bentonville, AR [email protected]
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
