Re: iSCSI: unsolicited data

[Paul Koning <ni1d@arrl.net>]

>>>>> "Julian" == Julian Satran <Julian_Satran@il.ibm.com> writes:

 Julian> Sandeep, I assume that TCP will not overcomit (this is also a
 Julian> commitment that is easy to predict and small in size).  iSCSI
 Julian> in this case is on the safe ground whatever it commits.

I assume TCP doesn't overcommit, but that doesn't help.  TCP worries
about TCP resources, and iSCSI has to worry about iSCSI resources.
You may be tight on iSCSI resources and still have TCP buffers.

A way to avoid that is to be aware of how TCP manages resources and
manage iSCSI resources "in parallel" and more conservatively.  If so,
then you are never "overcommitted" at the iSCSI level, and you need no
flow control in the first place.

The reason for having iSCSI level flow control is that the above may
not be true -- i.e., TCP may have resources when iSCSI doesn't.  

So those cases where iSCSI flow control adds value are also the cases
where lower layer flow control doesn't protect you from getting in
trouble if you overcommit.  In other words, if you want reliability,
don't overcommit at any layer.

      paul