ValerioZ <(E-Mail Removed)> wrote:
> Do someone know any document which explain "TCP segmentation offload"?
IIRC, Microsoft had some stuff on their website describing large send
- what was expected of drivers and the like. Beyond that, some
specifics of implementation in the NIC(s) would probably be in the
programmers manuals of the NICs supporting the functionality.
In a nutshell, the NIC says to the stack, "Hi, I know how to turn one
large TCP segment into several smaller TCP segments that I will send
on the wire. That way you don't have to make as many trips down the
protocol stack and will save CPU." And then the stack says something
like "Swell, here is a large segment. Please send it."
Of course, the devil is in the details - like making sure... that when
TSO is enabled, the TCP stack still complies with things like
slow-start at the beginning of a connection, and after a
retransmission timeout and all that sort of thing. Then there are
littler things - like what happens when you try to take a tcpdump
trace on a TSO-enabled NIC - btw, what happens there is the large send
becomes a line stating "IP bad-len 0" in the trace. Very helpful
indeed...
rick jones
--
The glass is neither half-empty nor half-full. The glass has a leak.
The real question is "Can it be patched?"
these opinions are mine, all mine; HP might not want them anyway...

feel free to post, OR email to raj in cup.hp.com but NOT BOTH...