RE: ISCSI: question about text command data

[Bart Crane <bcrane@iready.com>]

Title: RE: ISCSI: question about text command data

Because there is no guarantee that the terminating nul is actually there, you have to examine every character up to the last according to the data segment size.  Thus, "at least one" has negligible effect on performance.

Some implementations may keep their strings on 4- or 8-byte boundaries, zero-padded to a multiple of 4 bytes, use 2 fewer bits to represent their size, and dma them directly into the network adapter.  In this case, "only one" could impact performance negatively. 

The double-null can be handy, but it can also be viewed as a waste of a byte;  an "implied" terminating null exists at the character pointed to in the data segment at the offset specified by the size of the data segment.  And because that null is not guaranteed to be there, it may as well exist by implication only.

I vote for "at least one" null. 

-----Original Message-----
From: Buck Landry [mailto:blandry@Crossroads.com]
Sent: Tuesday, September 18, 2001 2:22 PM
To: ietf-ips
Subject: RE: ISCSI: question about text command data

The below might imply that I actually take a position on what should
appear in the data segment after a key-value pair.

I don't care much *what* goes in there, both approaches have good
reasoning; I would just like to see it specified as 'exactly one', or
'one or more', so there's no confusion.

If we decide on 'exactly one' .. (are leading NULLs allowed?) .. that
would seem to preclude using "double nulls" as Lee speaks of. (unless
julian somehow takes that into account in his wording)

regards,
Buck

-----Original Message-----
From: CONGDON,PAUL (HP-Roseville,ex1) [mailto:paul_congdon@hp.com]
Sent: Tuesday, September 18, 2001 10:11 AM
To: 'Steve Senum'; ietf-ips
Subject: RE: ISCSI: question about text command data

I disagree.  There is no reason why the spec shouldn't be very specific
about what behavior to take.  In the interest of simplicity, the spec
should
state 'exactly one null character' as Mark and Buck request.   If people
follow the spec, there won't be a need to invoke costly error response.

Paul

-----Original Message-----
From: Steve Senum [mailto:ssenum@cisco.com]
Sent: Monday, September 17, 2001 8:27 AM
To: ietf-ips
Subject: Re: ISCSI: question about text command data

Mark,

I agree with Julian on this issue.

Steve Senum

Mark Bakke wrote:
>
> Julian-
>
> Wouldn't it be simpler to just say "exactly one".  The last
> part of Buck's question mentioned that he didn't see why
> anyone would want more than one, and nobody responded saying
> they did.
>
> Thanks,
>
> Mark
>
> Julian Satran wrote:
> >
> > I've changed it the text to "at least one" to avoid errors hard to
list.
> >
> > Julo
> >
> > "Buck Landry" <blandry@Crossroads.com>@ece.cmu.edu on 13-09-2001
01:25:36
> >
> > Please respond to "Buck Landry" <blandry@Crossroads.com>
> >
> > Sent by:  owner-ips@ece.cmu.edu
> >
> > To:   <ips@ece.cmu.edu>
> > cc:
> > Subject:  ISCSI: question about text command data
> >
> > I have a small question about what separates the "key=value" pairs
in
> > the data segment of an iscsi text command.  On pg. 78 of the iscsi
v7-90
> > draft (2.10.5), it states:
> >
> > >>>
> > Every key=value pair (including the last or only pair) MUST be
followed
> > by null (0x00) delimiter.
> > <<<
> >
> > The question: is it legal to have *more* than one null char between
> > key=value pairs?  (no, I don't know why anybody would particularly
want
> > to do this.)
> >
> > Thanks,
> > buck
>
> --
> Mark A. Bakke
> Cisco Systems
> mbakke@cisco.com
> 763.398.1054