RE: TCP limitations (was Re: ISCSI: Urgent Flag requirement violates TCP.)

["Y P Cheng" <ycheng@advansys.com>]

> > The relationship of anonymous buffer size is really to 1/sqrt(p),
> >  and not the BW or the RTT per-se.
>
> Yes!, well put.
>
> Viewed another way: it is *really* hard to get TCP to sustain a very high
> data rate over a WAN.  So I question whether the design point of GB/s
> long-haul rates makes sense in practice.  If it doesn't, then the
> buffering
> requirements go way down, and it's not clear that failing to find framing
> for out-of-sequence segments is in fact all that expensive.
>
> 		Vern

Just because the old software TCP implementation could not sustain a very
high data rate over a WAN using TCP, it is not OK to conclude the design
point of GB/s with TCP not making sense in practice. In a couple of years,
the WAN and MAN and LAN will start to use 10 Gb/s connection.  The Internet
community must raise up to this challenge even if we must change the old TCP
implementations. I do believe it is possible for a TOE adapter to run in
slow mode with old TCP implementation and fast mode with another TOE
adapter.  We just have to make sure the iSCSI spec. does not prevent that.