TCP/UDP checksum offload

Does (will) io-net infrastructure support TCP/UDP “HW checksum offload”
that is embedded in ethernet chip such as 82559 or newer gige ethernet chip?

If not, how will QNX keep up to the performance/cpu usage in
Linux, NT, BSD world?


-Tony
Nokia Networks.

“Tony Lee” <tl_168168@hotmail.com> wrote in
news:a6941n$2ss$1@inn.qnx.com:

Does (will) io-net infrastructure support TCP/UDP “HW checksum offload”
that is embedded in ethernet chip such as 82559 or newer gige ethernet
chip?

If not, how will QNX keep up to the performance/cpu usage in
Linux, NT, BSD world?

AFAIK it’s being looked into - one issue I see is that there is more than
one CRC, and different chips do a different number of crc offloading’s. So
a negotiation on functionality needs to be done based on the chips feature
set.

\

Cheers,
Adam

QNX Software Systems Ltd.
[ amallory@qnx.com ]

With a PC, I always felt limited by the software available.
On Unix, I am limited only by my knowledge.
–Peter J. Schoenster <pschon@baste.magibox.net>

“Adam Mallory” <amallory@qnx.com> wrote in message
news:Xns91D26C5D7C23amalloryqnxcom@209.226.137.4

“Tony Lee” <> tl_168168@hotmail.com> > wrote in
news:a6941n$2ss$> 1@inn.qnx.com> :

Does (will) io-net infrastructure support TCP/UDP “HW checksum offload”
that is embedded in ethernet chip such as 82559 or newer gige ethernet
chip?

If not, how will QNX keep up to the performance/cpu usage in
Linux, NT, BSD world?

AFAIK it’s being looked into - one issue I see is that there is more than
one CRC, and different chips do a different number of crc offloading’s.
So
a negotiation on functionality needs to be done based on the chips feature
set.

The tcp/udp portion of io-net need to have APIs that let the driver to
tell them that they don’t need not to calculate the checksum.

From that point on, it is the individual driver issue.


-Tony Lee
Nokia Networks

“Tony Lee” <tl_168168@hotmail.com> wrote in
news:a75ahk$5lt$1@inn.qnx.com:

The tcp/udp portion of io-net need to have APIs that let the driver to
tell them that they don’t need not to calculate the checksum.

From that point on, it is the individual driver issue.

Sure, but as I mentioned, it’s not so trivial. io-net doesn’t make an
assumption that the underlying link layer is ethernet. So the design has
to take into consideration that offloading could be meaningless to the link
layer, or that the LL only allows certain types of offloading.

In the end, AFAIK, this is already in the works - I’m just stating the
obvious.

\

Cheers,
Adam

QNX Software Systems Ltd.
[ amallory@qnx.com ]

With a PC, I always felt limited by the software available.
On Unix, I am limited only by my knowledge.
–Peter J. Schoenster <pschon@baste.magibox.net>