RE: Avoiding deadlock in iSCSI

["John H Dowdy/Raleigh/IBM" <jdowdy@us.ibm.com>]

David Robinson wrote:
> Any TCP implementation that offers a total window size (sum of
> windows on all connections) greater than available buffer space
> is just plain broken, so this is not an issue.

Actually, I think this is the issue. Specifically, like airlines,
we would like to advertise a total window which is greater than the
available buffer space. Generally, this helps "keep the most seats
full" (i.e., utilize bandwidth the best).

I won't argue that this is "not broken" in design. However, I have
products in the field that do just that, and no one has thought them
to be broken, because never have all the tcp receive buffers been "full"
at the same time.

Having said that, I do agree with David that when you advertise a
TCP window, you are implying that you can buffer that much data.

JD.