Browse Prior Art Database

Recovering from SCSI Bus Errors on Sequential Access Devices

IP.com Disclosure Number: IPCOM000111396D
Original Publication Date: 1994-Feb-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 4 page(s) / 143K

Publishing Venue

IBM

Related People

Chartrand, DM: AUTHOR

Abstract

A general method for recovering from interface errors on sequential access (tape) SCSI devices is disclosed. The recovery is independent of knowing device-positioning characteristics after an error occurs, and can be implemented to operate properly regardless of the exact nature of the error which occurred.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 36% of the total text.

Recovering from SCSI Bus Errors on Sequential Access Devices

      A general method for recovering from interface errors on
sequential access (tape) SCSI devices is disclosed.  The recovery is
independent of knowing device-positioning characteristics after an
error occurs, and can be implemented to operate properly regardless
of the exact nature of the error which occurred.

      Effectively recovering from all types of SCSI errors on
sequential access devices is very difficult.  The problems
encountered generally relate to maintaining and knowing the current
device position so that the recovery is able to reposition properly
after the error and retry the command.  In this respect,
random-access devices have simpler and more effective error-recovery
procedures, since the command to the device entails an absolute
position, and can generally be re-issued without concern for the
state of the device after the error.  Previous forms of recovery for
sequential access devices have been based on knowing the exact
circumstance of the error, the ending status of the command, and the
device specific positioning characteristics with respect to the
ending status and command in progress.  The resulting recovery
procedures were complex, fragile, and not easily verifiable to be
correct under all circumstances.

      There are two different aspects to a generic error-recovery
procedure for sequential access SCSI devices.  First, there are
certain device requirements to be met in order to enable the error
recovery.  Next, the host system (or I/O processor) must implement
the generic error-recovery procedure.  Both of these items are
described below.

      Device Requirements

In order to enable error recovery for a large set of SCSI bus errors,
the device must be designed to meet the following requirements, in
addition to the normally imposed requirements of the SCSI-II
standard.  These requirements are not a violation of the SCSI-II
architecture, but are a definition of implementation characteristics
for the device for options which are part of the SCSI-II standard.

1.  Upon receiving a SCSI bus-reset or device-reset message, the
    device must, if possible, just clear out the information
    corresponding to the command(s) which are currently executing.
    These are the commands which have been identified but have not
    sent command complete.  The buffer for any commands which have
    sent command complete is to be maintained.  The current device
    state is to be maintained.  The device is required to release the
    bus immediately (as identified by SCSI-II).

          If this is not implemented by the device, the general
    error-recovery procedure will still operate correctly.  However,
    recovery from error scenarios involving a reset may be limited
    unnecessarily.

2.  The device must generate a unit attention condition after all
    resets.  The unit attention cond...