Browse Prior Art Database

Isolating the Source of Small Computer System Interface Bus Hang Error at Run-Time

IP.com Disclosure Number: IPCOM000117994D
Original Publication Date: 1996-Aug-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 54K

Publishing Venue

IBM

Related People

Fore, RC: AUTHOR [+3]

Abstract

Disclosed is a method via which a Small Computer System Interface (SCSI) controller may exploit standard SCSI device error-reporting mechanisms which initiate a run-time diagnostic sequence designed to help identify the cause of SCSI bus hang errors.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 57% of the total text.

Isolating the Source of Small Computer System Interface Bus Hang
Error at Run-Time

      Disclosed is a method via which a Small Computer System
Interface (SCSI) controller may exploit standard SCSI device
error-reporting mechanisms which initiate a run-time diagnostic
sequence designed to help identify the cause of SCSI bus hang errors.

      Since SCSI devices communicate over a set of shared data and
control signals, it is possible for one errant device to "hang" the
bus and prevent any other attached device from communicating.  This
behavior can make it difficult to isolate the faulty device when a
SCSI bus hang  occurs.  For example, consider a typical Redundant
Array of Independent  Disks (RAID) configuration.  In a RAID
application, the SCSI initiator  device is typically processing
concurrent Input/Output (I/O) requests to  a set of attached target
devices.  Now suppose that one of those targets  encounters an error
and hangs the SCSI control lines (and thereby prevents any of the
other targets from re-connecting to finish an outstanding I/O
request).  Under these circumstances, the SCSI controller  will not
detect an error unless a background command timeout function is
provided to alert the controller that one of the outstanding I/O
requests has timed-out.  Furthermore, the first request to time-out
may or may not  be associated with the errant target device, which
means that the initiator must use some other method to attempt to
isolate the...