RE: iscsi: unsolicited data question

[Bill Studenmund <wrstuden@wasabisystems.com>]

On Wed, 12 Jun 2002, Dennis Young wrote:

> Are you saying that, for a session that has InitialR2T=No in effect, the
> initiator
> must send all its data as unsolicited first, up to the amount negotiated in
> FirstBurstSize, before it waits for a R2T from the target?

No. See the note I sent which crossed this one in the mail.

If the initiator doesn't set the F bit in the command PDU (and
InitialR2T=No), then it indicates that it will act as if it has already
received an R2T for up to FirstBurstSize worth of the data.

However the initiator can CHOOSE to set the F bit, and then it will send
no more data until it receives an R2T.

> Can you shed some light on why we need unsolicited Data-out PDU when there
> is ImmediateData, seems like they both serve the same purpose, having both
> of
> them only make the spec more complex.

How big do you think FirstBurstSize and MaxRecvPDUDataSize will be? The
defaults are 64k and 8k respectively. So with InitialR2T=No and default
numbers, 8 PDUs worth of data can be sent w/o need for an R2T, whereas
with just immediate data, only 1 PDUs worth of data won't need to wait.

While I don't have a histogram of typical i/o sizes (and even that would
be OS and task-set specific), I expect that a good number will be in the
8k to 64k range. So unsolicited data permit writes of those sizes to not
need to wait for the target to send an R2T.

Take care,

Bill