FCP-2 SRR for lost Write Data IUs.

Santosh Rao santoshr at cup.hp.com
Fri Mar 29 17:43:48 PST 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* Santosh Rao <santoshr at cup.hp.com>
*
Hello,

I have some questions regarding FCP-2 sequence level error recovery for
lost Write Data IU case. (described in FCP-2 rev 07a Sections 12.4.1.6).

Any clarifications on the below questions would be appreciated.

1) Initiator detects that a write data iu in the middle of an outbound
data sequence it sent was lost.
 
How should an initiator detect what is the relative offset to be used in
the SRR when attempting to recover from such a scenario ? The data
transfer count in REC ACC only indicates the number of bytes the target
has received so far, and not which data IU was lost. Hence, the
initiator will have to send an SRR with the same relative offset as the
data_ro in the last XFER_RDY it received.

If an initiator does not save a copy of the last FCP_XFER_RDY request it
receives each time, then, it does not have the relative offset
information for the SRR. (XFER_RDY processing may be in
firmware/hardware and existing FCP based hardware/firmware
implementations may not have this data, in which case, drivers do not
know what was the last data_ro XFER_RDY it received.)

2) If an existing initiator implementation does not have this facility
to maintain a copy of the last XFER_RDY it responded to, then, can it
send a SRR with a relative offset of 0 ? (This is allowed when initiator
detects a lost read data iu, but there's no mention of whether it is
allowed for a lost write data iu.)

3) Section 12.4.1.6 states that on receiving a SRR for a given relative
offset, the target must discard the sequence in error. In the case of
non-idempotent targets (ex : tape devices), if an SRR was received for a
relative offset of 0, and the previous sequences were already commited
to the tape medium, then, what is the target response ?

4) Does support for sequence level recovery imply that FCP-2 tape
devices that support SRR, buffer received data sequences and not commit
them to the tape medium until the exchange completes and FCP_CONF is
received ?

If not, how do they deal with an initiator that will always send SRR(ro
= 0) on any write data iu loss ? Should the behaviour in this case be
explicitly called out in FCP-2 spec ? (I can't find anything in the spec
that speaks about what non-idempotent devices should do to support SRR.)

5) Can any existing FCP-2 tape product implementors comment on how their
implementations would handle scenarios (3) & (4) above ?

Thanks,
Santosh


-- 
##################################
Santosh Rao
Software Design Engineer,
HP-UX iSCSI Driver Team,
Hewlett Packard, Cupertino.
email : santoshr at cup.hp.com
Phone : 408-447-3751
##################################
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list