SCSI, Networks & Sense

Jim McGrath jmcgrath at qntm.com
Wed Mar 29 16:28:17 PST 1995


        Reply to:   RE>>SCSI, Networks & Sense Data


Ken wrote:

...

No one would expect ECC codes to be dealt with by ULPs nor drive
positioning errors. Error recovery is best left to the lowest level
possible. But what about error statistics and regression analysis? Is a
link degrading? Where was the unit in a lengthy operation when it failed?
These and other minutia regarding errors may be of interest to upper layers
if the information is preserved. The type of information preserved would
only have relevance when considered in the context of the task being
attempted, thus the upper layer would be the place for intrepretation. I
agree that link specific information like message parameters and such are
of no concern to the ULP and would have no meaning outside of the link
structure. However, there might be plenty of SCSI command information
concerning a fault that the Sense data could supply.

My reply:

I would like to suggest that if people really want SCSI sense information to
essentially hang around indefinately then a SCSI LOG SENSE page should be
defined to have a (rather long) circular list of error entries.  Each error
entry would contain command identification information (queue tag, initiator,
LUN, CDB bytes) and status (possibly all of the sense data).  This could be
done in a 64 byte entry easily.  Then if people want old sense data than they
can get it there.  I can store it on a disk, so can offer a lot of it (64K =
1000 error events).

In general, do not use the sense return mechanism for this - use the LOG
mechanism for it (that is what it was designed for).

Jim







More information about the T10 mailing list