This message is from the T13 list server.

As I asked in the previous email, I don't see a proposal for the
WRITE DMA QUEUED FUA command. But looking at the ATA/ATAPI-7
description (both the general description of O/Q and the command
description) I don't see a problem. Basically this command is a O/Q
command (so it does not abort the outstanding command queue) but it
is a command that is not allowed to release (so it executes like an
non-O/Q command). Yes, that may be a big performance problem but that
is the price you pay.

HOWEVER, there is a big problem if the host is dumb and has
outstanding O/Q commands, read and/or write, that access the same
sectors that another O/Q command accesses (including the WRITE DMA
QUEUED FUA command). Again, unless I missed something, I think
ATA/ATAPI-x has nothing to say about the action and/or data integrity
of O/Q commands that access the same sectors. I think when O/Q was
added for hard disk drives T13 assumed this was the host's problem.

But the O/Q EXT commands have yet another problem (that I have asked
about before). If these commands are used with really large transfer
lengths (remember that and EXT command can tranfer up to 65536
secotors or about 33Mbytes of data) these commands have a real
performance problem in that only one release/service cycle is allowed
(and that must be before any data is tranferred). The ATAPI DMA O/Q
option of the PACKET command allows a device to release/service
multiple times (before and during the data transfer). O/Q EXT
commands should allow the same for hard disk drives.

Now the "real" question... Does any of this O/Q stuff matter? Doesn't
SATA have something that replaces all this old PATA O/Q stuff?

Hale



*** Hale Landis *** www.ata-atapi.com ***



Reply via email to