RE: Review of the 07 draft

["Julian Satran" <Julian_Satran@il.ibm.com>]

Robert,

Snippets and comments in text - Julo





Section 1.2.6
...

+++ Logout has also history. We started without a logout but realized
soon
that we will need a synchronized logout to enable command allegiance
change
for recovery (that is possible only after logout). Besides it is a
simple
and painless way to do a clean shutdown of a connection. As for a target
only logout that is what the Asynch Message does (after some timeouts)
Please keep in mind that there is also a fine interplay between logout
and
recovery and that both initiator and target may decide to start some
recovery action+++

[Bob Griswold]  I see the need for the logout to be as quick and as
clean as possible.  If one assumes the Initiator and Target have agreed
on some form of possible recovery scenario, then having the target spend
cycles trying to deliver some recovery information in response to a
Logout seems reasonable; but I am not confident divergent iSCSI
platforms could come to such an agreement, short of it being defined in
one of the iSCSI specifications.

>>>> Logout brings commands associted with the coonection in a state in
which they can change allegiance. This is specified in 7 and recovery
appendix <<<<<<






As for not  returning anything this is acceptable for Target Cold Reset
-
where the TCP connections are teared down but not for those that leave
the
connections standing +++

[Bob Griswold]  I have reread this comment, but cannot understand what
you are saying.

>>>> Cold reset tears down TCP connections <<<<

Section 2.8.5
...

+++ We think that hex will be the preferred encoding used but sometimes
people will copy decimal numbers. Once you have to have a decimal
conversion routine it makes little sense to force hexadecimal only by
dictate
+++

[Bob Griswold]  Well, if you assume that someone somewhere will have to
do the conversion of decimal to hex, then the burden has to live
somewhere, right?  If the wire is going to be exposed to two different
encoding schemes, then how is a lowly target going to know if its
getting decimal or hex values?  Will there be some marker, key=value
system, or something that will allow the receiver of an ambiguous data
type to know what it is getting?
>>>>> Syntat - Hexadecimal start with 0x <<<<<<<<


keeps sending what it believes to be valid PDUs, but appears to the
target to be invalid, then where and how does the initiator learn about
what is going on at the target level?

Section 2.16.1
...
+++ The target should request logout if it supports connection recovery
but
not
data snack +++

[Bob Griswold]  So is that information (supporting recovery but no data
snack) inclusive in the discovery information?

>>> That will be negotiated recovery level <<<<


Section 7.1c
This description of the use of the Retry bit seems to put the onus of
interpretation of the Retry bit on the target for command replay.  Is it
at all possible for a Replay request to get issued, within the scope of
an initiator delaying its acknowledgement, such that the command
actually did get completed, but the Retry bit was set too soon?  Would
one assume that an initiator just avoids doing something like this?

+++ I will come back to this later. +++

[Bob Griswold]  Great, I am looking forward to it.

>>> We will at tiny encoded field to say what the retry means in a day or
two <<<


Thanks,
Bob

Robert Griswold
Technologist
Crossroads Systems, Inc.
512-928-7272