OPTCD=C chained scheduling goes all the way back to OS/360, and was not limited 
to DASD. Starting with SAM-E, a more efficient type of chained scheduling was 
automatic for DASD. BSAM and QSAM chain new requests to the current channel 
program and reschedule them if channel program terminates too soon. There was 
documentation, e.g., MVS/Extended Architecture SAM Logic Data Facility Product 
5665-XA2 Version 2  Release 1.0, LY26-3967-0,  
http://bitsavers.org/pdf/ibm/370/MVS_XA/DFP_Version_2/LY26-3967-0_MVS_XA_DFP_Version_2_Release_1.0_SAM_Logic_Apr1985.pdf,
 but I'm not aware of any current documentation.


--
Shmuel (Seymour J.) Metz
http://mason.gmu.edu/~smetz3

________________________________________
From: IBM Mainframe Discussion List [[email protected]] on behalf of 
Joel C. Ewing [[email protected]]
Sent: Sunday, March 21, 2021 11:08 AM
To: [email protected]
Subject: Re: Overlapped I/O completion

Believe it.

Even before emulated DASD,  MVS QSAM would read multiple blocks with a
single channel program to eliminate rotational delays on native DASD and
do anticipatory reads for the next set of buffers even while you were
still processing records in blocks from the previous read, as long as
you specified enough buffers.  You could prove this with native
3380/3390 by measuring real execution time and increasing the number of
buffers.  Once physical and virtual memory space were not a limiting
factor, the recommendation was to specify enough buffers to at least do
full-track reads as a minimum for all sequential files.  Not sure what
the upper bound was -- probably the max number of blocks per physical
read was 1 cylinder, but could have been less.  The real time effects
were incredibly dramatic if you found an application with smaller block
size and an inadequate buffer specification and fixed the JCL.

Pretty sure with emulated DASD that QSAM is smart enough to communicate
to the DASD subsystem that it is doing pure sequential access and needs
to pre-stage data into cache in the DASD subsystem.   Since BSAM doesn't
have to be purely sequential, it may not be as good at communicating
future intent to the DASD Subsystem.   Obviously with emulated DASD, the
z/OS concepts of device track and cylinder boundaries don't really
apply, but the DASD subsystem can see the sequential access intent and
translate that into whatever access maximizes data transfer on its
internal physical disks.

So yes, I would expect properly tuned QSAM I/O to out perform even an
incredibly well designed overlapped BSAM I/O access strategy on pure
sequential access, and with much less work.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to