Read Long Conundrum
jmcgrath at mail.qntm.com
Mon Oct 3 13:59:10 PDT 1994
Reply to: RE>Read Long Conundrum
... I am one who would feel agony if this command
were deleted. It is not uncommon for drives to not
report data errors which they were able to correct
with on-the-fly ECC, no matter how the PER bit is set.
It is difficult to detect that a drive is doing this "favor"
for you without the Read/Write Long commands.
Unfortunately, no interpretation of READ LONG/WRITE LONG gets
you what you want. Quantum considers single bit errors that
are corrected on the fly to be part of the read channel, not the
controller. Channel errors are not governed by SCSI, and so need
never be reported as ECC errors. In this we are following the usual
industry practice. Indeed, with PRML channels the concept of
an uncorrected (by any means) data stream is technically meaningless.
There are simply corrections controlled by the read channel (which
SCSI does not control) and those by the controller (which
SCSI does control). And since you may be talking about the
same chip in both cases, the dividing line between the two
is vendor unique.
I would strongly suggest that these commands be deleted or
made vendor unique. Testing of single blocks is largely
meaningless. Testing of large blocks provides data that is
better provided via the LOG SENSE/LOG SELECT commands.
And forget trying to get channel performance data via SCSI -
that is not what the interface was designed to do, and given
the vendor unique nature of the data it is almost impossible
to use it anyway.
More information about the T10