Browse Prior Art Database

Correction of Channel Measurement Data Caused by Control Unit Queuing

IP.com Disclosure Number: IPCOM000120235D
Original Publication Date: 1991-Apr-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 4 page(s) / 193K

Publishing Venue

IBM

Related People

Fredericks, KJ: AUTHOR [+5]

Abstract

Disclosed are three methods for the program to acquire additional measurement information that is the result of control unit queuing.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 31% of the total text.

Correction of Channel Measurement Data Caused by Control Unit Queuing

      Disclosed are three methods for the program to acquire
additional measurement information that is the result of control unit
queuing.

      Typically on XA/370 and ESA/370* systems queuing of I/O
operations is performed at the channel subsystem. In a shared
environment this implementation is sensitive to the processor's
channel timing.  A direct access storage device (DASD) can be locked
out from one or more processors because of the channel timing being
faster on a different processor. In addition, this implementation
causes extensive "device- busy"/"device-not-busy" transactions on the
I/O interface, resulting in non-productive work.

      A future DASD product performs queuing of I/O operations at the
control unit.  The queuing function is implemented by a large number
of Task Control Blocks (TCBs) and the associated control
logic/microcode.  When a chain of I/O commands is initiated, the
chain is tied to a TCB.  If all the TCBs are in use, then
control-unit-busy status is indicated to the channel.  Since the
number of TCBs is large, the occurrence of this condition should be
infrequent.

      The following device-availability verification is made by the
control unit with the indicated outcomes:
1.   Device is available and not Reserved
        Accept the chain and start processing.
2.   Device is not available and not Reserved
           Another system is using this device, but that system will
eventually complete its I/O operation.
           Accept the chain and its start-up commands (Seek, Set File
Mask, Set Sector, etc.) and then disconnect until the device becomes
available.

      This is the most useful case for control-unit queuing, because
it prevents a device lock-out to a particular system. This prevention
of device lock-out is possible, since all of the systems that request
service for a device are lined up in a first-in, first-out (FIFO)
queue. The system that is using the device will complete its I/O
operation and may attempt to get back to the device as soon as it
can, but the system will end up at the bottom of the FIFO queue.
3.   Device is Reserved
           Device busy is indicated in the status until the other
system holding the Reserve completes its series of I/O operation and
issues a Release command.
4.   Device not available and not Reserved and the new chain to be
initiated starts with a Reserve Command
           Command retry is indicated in the status to the channel.
      The control unit indicates device busy in the status to all
subsequent requests for this device.  The control unit waits for the
preceding chains to complete and the queue to drain (empty), then
reconnects to the channel with the Reserve and processes the chain.

      When a control unit queues an I/O operation, the channel's
accumulated performance information, reported by...