RE: sector alignment for DataOut PDUs?

["Rod Harrison" <rod.harrison@windriver.com>]

Nick,

	This makes sense for unsolicited data but it doesn't stop the
initiator sending more than one "weird" sized DATA-OUT PDU in
response to an R2T, even if the R2T is for a multiple of 512
bytes.

	Strengthening the wording to make sending "full" PDUs a
requirement is not really practical in the post v08 world where
we don't have a negotiated PDU size. It would have made sense
when the max PDU size was agreed upon but given the current
scheme of allowing the receiver to name an arbitrary sized
maximum receive PDU it would place an unreasonable burden on the
initiator.

	- Rod

-----Original Message-----
From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On
Behalf Of
Martin, Nick
Sent: Wednesday, February 27, 2002 12:14 AM
To: Paul Koning; ips@ece.cmu.edu
Subject: RE: sector alignment for DataOut PDUs?


Paul,

For all data carrying PDUs except the last in a sequence, the
sender is
expected to send maximum sized PDU allowed.  When the negotiated
maximum
is a multiple of 512, this effect is what you request.

I thought this was a requirement, but I did not find it as such
in draft
10.  I did find this:

: 8.5 Unsolicited Data and Performance
: Unsolicited data on write are meant to reduce the effect of
latency on
: throughput (no R2T is needed to start sending data). In
addition,
: immediate data are meant to reduce the protocol overhead (both
bandwidth
: and execution time).
: However, negotiating an amount of unsolicited data for writes
and
: sending less than the negotiated amount when the total data
amount to
: be sent by a command is larger than the negotiated amount may
negatively
: impact performance and may not be supported by all the targets.

This is a warning that an initiator sending less than the
negotiated
maximum when the expected data transfer is greater than the
maximum a)
may reduce performance and b) may not be supported by all
targets.

IMHO, it makes more sense to include stronger wording encouraging
maximum negotiated length transfers rather than to add another
parameter
requiring PDU breaks on different boundaries.

If such initiator behavior may not be supported by all targets,
then the
initiators SHOULD NOT do it.

A disk target which can not handle such behavior is forgiven in
advance
;).

Thanks,
Nick


> -----Original Message-----
> From: Paul Koning [mailto:ni1d@arrl.net]
> Sent: Tuesday, February 26, 2002 3:51 PM
> To: ips@ece.cmu.edu
> Subject: sector alignment for DataOut PDUs?
>
>
> In the past, a number of login parameters were specified as
multiples
> of 512 bytes, which makes a lot of sense for disk targets (but
not so
> much for tape targets).  With the spec as it stands now, you
can
> negotiate pretty much arbitrary burst sizes etc.  And in any
case, the
> sender can pick "strange" PDU sizes even if the negotiated
sizes are
> multiples of 512, because after all those are limits, not
required
> sizes.
>
> It would be very attractive for disk targets to be able to
specify
> that they require DataOut PDUs to be multiples of 512 bytes in
length.
> That way, any PDU would correspond exactly to one or more
sectors,
> rather than potentially having several PDUs straddling sector
> boundaries as is currently permitted.  This could be done by a
Boolean
> key (512 byte alignment, yes or no).  (An integer key would
allow
> other values than 512, but I'm not sure that such flexibility
is a
> whole lot more useful.)
>
> Basically, the effect of this feature would be to tell the
initiator
> that it must send DataOut PDUs and immediate data whose length
is
> always a multiple of 512.  Obviously, targets can't ask for
that
> unless the devices on that target already come with such a
limitation.
>
>        paul
>
>