RE: iSCSI: Flow Control

[Joshua Tseng <jtseng@NishanSystems.com>]

Hi David,

>Could someone please clearly state the problem that
>iSCSI "flow control" is attempting to solve.  Given
>the reliable stream nature of TCP with its flow
>control, what problem exists? Also with an iSCSI
>layer flow control will we not also be victim to some
>nasty hysteresis effects due to TCP's flow control as
>well?
>
>	-david
	
As I understand it, TCP should not have any intelligence
of the iSCSI layer above it.  Thus, TCP does not know if
the oncoming data is an iSCSI data PDU or a command PDU.
If it is receiving a long stream of data, TCP can't tell
if it's 1 short command PDU and a very long data PDU, or
100 short command PDU's. The latter case could overwhelm the
SCSI command queue, but still be within the TCP window of
the receiving node.  That is, unless you want iSCSI to talk
to TCP to tell it to close the window.  But that would
violate layering principles, I believe.

Josh