ACA handling for "temporary initiators"

[Tom Wicklund]

I agree with the other responses from Jeff Williams, Bob Snively, and
Jim McGrath.  The temporary initiator should handle the ACA condition.
If initiator A has to handle it, A should send commands to B, and B
pass them through to C as required to resolve the condition.

Bridge controllers do this all the time.  A "traditional" RAID
controller consisting of a SCSI host port and SCSI device ports deals
with all device errors internally, shielding the initiator.

There is also a case not shown in the Seagate document, but perfectly
reasonable.  Assuming dual port SCSI drives, temporary initiator B
might issue a command to target C on the alternate SCSI bus (thus
spreading the load around).  In this case initiator A may not be able
to talk to target C, or if so only through a more complicated connection.

I understand the reason for Mr. Houlder's request from since I can
envision error recovery actions which require sending commands to all
devices in an array, which might be difficult when funneling all
requests through a single device, but complicating ACA and potentially
making it less reliable doesn't the best option either.