EMDP
It is said that EMDP is useful for offering data that is
being buffered from multiple
non-synchronized sources (like RAIDed disk drives) as
quickly as possible.
Whichever blocks of data are accessible first are
delivered first, reducing average
latency significantly.
I believe that, even if the mechanism is not yet used,
high performance block
storage devices may eventually like this function and
that it should not be
removed.
data
overlay
It
is said that data overlay may be useful for immediate delivery of
buffered
data
that is later found to be incorrect and must be modified. This behavior
may
appear on tape drives. The concern about
problems with verifying that all the
data
has been delivered to the correct location may be manageable
through
the
improved data integrity features being provided now in SCSI. It may
also
be
managed through various types of software managed check
sums.
I
believe that, even if the mechanism is not yet used, this
architectural
capability should not be removed from
FCP-4.
I
don't believe that any harm comes from leaving both of these in FCP-4. In
the
future, there may be significant benefits from
exploiting these functions. At present,
they
are clearly optional and may be left out of any implementation. Usage of
these
optional functions is negotiated early in the
initialization cycle of the SCSI devices.
Both
have default indicators of '0', making it easy to perform that negotiation,
even
if the
SCSI device is not aware of the possible existence of such
capabilities.
Bob
Howdy,
At the last FCP-4
working group meeting the use of Enable Modify Data Pointers (i.e., data
overlay) was questioned. I myself have never see the use of data overlay in the
FCP realm (other than via FCP-2+ error recovery which is strictly not considered
data overlay).
Please respond if
you object to obsoleting the EMDP bit and data overlay (other than error
recovery data overlay) in FCP-4 along with your reasoning.
Thanks...Dave
(no disclaimer)