Re: Command Reference Number in the iSCSI draft

[julian_satran@il.ibm.com]

Pierre,

CmdRN is supposed to be maintained across the adapters (a central counter)
and either handed by the adapter driver to the adapter or by a
read-modify-store
by the adapters. For outgoing data blocks there is no need to increment
CmdRN
as those can be always replayed from host memory either all or through RTT.

If the initiator sends without RTT and wants to recover by sending from
where it
stopped it has to keep track of data sent by himself and rely on the target
to request missing pieces.

I however wonder if in would not be wise to give-up on this optimization
and
have the initiator resend all data at recovery.

Regards,
Julo

Pierre Labat <pierre_labat@hp.com> on 22/08/2000 01:53:10

Please respond to Pierre Labat <pierre_labat@hp.com>

To:   ips@ece.cmu.edu
cc:    (bcc: Julian Satran/Haifa/IBM)
Subject:  Command Reference Number in the iSCSI draft




Hello,

A question about the CmdRN.
How does the CmdRM must handled in the data PDUs
in the case the initiator can't generate it?
The initiator can't generate it for example in the case of
several TCP connexions in a same session, each one using
a different adapter.
Which value to put in this field in this case, and how
to say to the target that it is meaningless?


Regards,

Pierre