Browse Prior Art Database

Purge Halt with Delay

IP.com Disclosure Number: IPCOM000104972D
Original Publication Date: 1993-Jun-01
Included in the Prior Art Database: 2005-Mar-19
Document File: 2 page(s) / 46K

Publishing Venue

IBM

Related People

Redding, I: AUTHOR [+2]

Abstract

This article describes a method for efficiently purging I/O operations from an operating system while preventing data corruption.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 58% of the total text.

Purge Halt with Delay

      This article describes a method for efficiently purging I/O
operations from an operating system while preventing data corruption.

Purging I/O operations associated with a unit of work is performed in
three steps:

o   Locate and dequeue all pending I/O operations that are queued but
    have not been started.
o   Attempt to withdraw started I/O requests (write operations for
    disk) from the channel subsystem before they become active at the
    device.
o   Wait for active write operations to disk to finish or x seconds,
    whichever comes first.
o   If waiting was required, then terminate operations that did not
    finish.

      In order to prevent corrupting data when a write command is
active on a disk device, a delay is introduced before actually
terminating the active I/O operation.  The delay must only last as
long as the operation is active, or N seconds, whichever comes first.
At the same time while waiting for the active I/O to finish, new
requests need to be prevented from being issued and queued requests
need to be prevented from being started.

      Note that the initial MVS implementation chose to use 2 seconds
as the value of N.  This value was arbitrary.  However, a
large-enough value needs to be picked that allows for typical disk
I/O requests enough time to finish but not so much that it would
adversely affect the systems operation during error recovery (i.e.,
trigger a take-over by anot...