TCP/IP Offload Engine

[Michael E Brown]

On Thu, May 03, 2007 at 01:01:53PM -0400, Hank wrote:
> I felt it was slanted due to biased comments like:

Ok, my Dell hat is off. Standard disclaimers apply.

After reading that page, I see that it is *remarkably* unbiased, when
compared to the absolute flamage I've seen on linux-{net,kernel}.

> "In order to configure a TOE NIC, hardware-specific tools are usually
> required. This dramatically increases support costs. "
> 
> Instead of something more balanced like "This may increase support costs"
> especially when they just said "tools are *usually* required"... or just
> drop the "dramatically".   For instance, if I have a room full of the same
> TOE NICs, and I use the same (vendor/hardware specific) tool to configure
> them (instead of linux tools), where's the "dramatically increased support
> costs"?  Nothing like a good old over generalization to get a point across.

So, instead of training your support personel to 'edit
/etc/sysconfig/network-scripts/ifcfg-ethX' to configure network, they
have to 1) is this a TOE system 2) if TOE, run vendor scripts, 3) forget
about "features X/Y/Z" because they arent supported on TOE. (packet
filtering/scheduling, etc)

Also, I assume you just bought your room full of servers and intend to
never buy another one ever again? Because three years later when you go
to upgrade your systems, that old TOE nic? Right, it isnt available
anymore. You have to upgrade to the newer one. Oh, and guess what? Yup,
it has a new configuration utility.

> ..or the statement "refer users to the vendor -- who in all likelihood cares
> more about non-Linux operating systems".. is also a very biased
> statement/conjecture/opinion.

Biased? Coming from a group of folks who have had to forcibly reverse
engineer whole families of network chips because the vendors refused to
release specs? (*cough*Broadcom*cough*) I'll go with the open source
guys on this one...

Broadcom now supports the TG3 driver for their wired gig-e stuff. This
is only because it was forced down their throats. They originally wrote
the bcm57xx driver, and only after having the open source guys eat their
lunch with the open source driver (faster, lower overhead, better
all-around performance) and having massive user (and OEM) backlash did
they decide to switch to TG3.

> And one more: "Supporting TOE requires *massive*, heavily invasive hooks
> into the network stack.".   It's just a little too dramatic of a comment
> there.  We get it.

Not dramatic if it is true. Have you actually seen some of the proposed
vendor patches to support TOE? They basically let any vendor say "Here
is a binary-only blob I need you to load if you want to use our latest
nic. Oh, by the way, it completely replaces the entire Linux network
stack."

> 
> It's clear that whoever wrote it had an axe to grind against TOE.

Or... it is clear that the people who wrote it are some of the primary
developers of the Linux kernel network stack and *know* *what* *they*
*are* *talking* *about*. Most of the information on that page was taken
directly from the primary authors of the Linux network stack from
mailing list discussions.

There are better ways to make the network go faster, and these have been
discussed on the linux-net mailing lists. TOE is a workaround to fix the
problems in one specific OS's slow network stack and is not a
generalized solution.
--
Michael
(not speaking for Dell)


> 
> -Hank
> 
> 
> On 5/3/07, Kuba Ober <kuba at mareimbrium.org> wrote:
> >
> >On Thursday 03 May 2007, Hank wrote:
> >> Here's good (but somewhat slanted) discussion why TOE is not supported
> >on
> >> Linux:
> >>
> >> http://linux-net.osdl.org/index.php/TOE
> >
> >It's pretty much dead-on. I can't say it's slanted.

> _______________________________________________
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> http://lists.us.dell.com/mailman/listinfo/linux-poweredge
> Please read the FAQ at http://lists.us.dell.com/faq