Re: iSCSI: Why is ACA optional?

[Ralph Weber <ralphoweber@compuserve.com>]

Costa,

} Some uses of disks require ordering of writes to maintain recoverability.
}
} Getting rid of ACA requires these applications to use stop-and-wait to
} ensure that updates are ordered to disk in a recoverable fashion.

I am not trying to get rid of ACA.  I am only explaining why T10 keeps
ACA optional in it's standards.  Specifically, the needs of "some uses
of disks" are not sufficient to make ACA support mandatory in T10
standards.

This is not to say that a T10 transport protocol standard is allowed
to omit specification of how ACA is optionally implemented by devices
using that transport protocol, each T10 transport protocol is required
to contain a description of its ACA implementation (but they all note
that implementation of ACA is optional).

I'm not even saying that iSCSI should make ACA optional.  The way IETF
treat optional elements of an RFC is very different from the T10 handling
of optional features.

As best as I understand it, an optional iSCSI ACA might not get implemented
by enough developers and thus get cut out of the final iSCSI standard.
Such an action would make iSCSI non-conformant with SAM-2 (a stated goal).
If Julian the other IETF process experts decide that the best way to have
any ACA is to make it mandatory, I'm not going to be a party pooper.

However, iSCSI will not get to reference T10 as its reason for making
ACA mandatory.  It is also useful to note that very few initiators set
the NACA bit in the CDB and if the NACA bit isn't 1 ACA doesn't happen.
So, there could be a lot of mandatory iSCSI code written that almost
never gets used.

Thanks.

Ralph Weber

P.S. I hope you don't mind me posting your comments to the reflector.