From The Linux Foundation

Some of the language in this document presupposes too much.

"A TOE net stack is closed source firmware" should read: "At this time, all the TOE net stacks that we know of use closed source firmware.".

"TOE is a black box" should read: "All the existing TOEs that we know of are black boxes".

And so on.

Now suppose a vendor changes to using open-source firmware for its NIC and TOE. Would the Linux developers then consider officially supporting it? The answer should be made clear.

If the answer is No, then no vendor will be motivated to open up its TOE.

But if the answer is a Yes or a Likely Yes, then this document should say so. This would provide some inducement to NIC vendors to switch to open source.

Rahul

Contents 1 Advocacy 1.1 Security updates 1.2 Point-in-time solution 1.3 Different network behavior 1.4 Performance 1.5 Hardware-specific limits 1.6 Resource-based DOS 1.7 RFC compliance 1.8 Linux features 1.9 Requires vendor-specific tools 1.10 Poor user support 1.11 Short term kernel maintenance 1.12 Long term user support 1.13 Long term kernel maintenance 1.14 Eliminates global system view

Advocacy

The bulk of the Net:TOE article seems to be value/advocacy style argumentation rather than data-based one. This may not be sufficient to convince those whose opinions don't already match those of the authors. For example:

Security updates

This would be less FUDdy if links were provided to actual TOE firmware security problems.

Point-in-time solution

All this says is that TOE may be faster on a given piece hardware. That's hardly an argument against it.

Different network behavior

Fair enough, maybe, for those for whom linux network system idiosyncrasies form valuable diagnostics.

Performance

Substantiation of the claim that connection-intensive protocols cause a net TOE performance deficit would be useful.

Hardware-specific limits

Substantiation of actual net harm please. The sentence makes it sound that TOE+fallback may be slower than TOE, but that's irrelevant to whether TOE is faster than non-TOE.

Resource-based DOS

Substantiation of actual net harm please, same as above. IOW how are people

• worse off* because of TOE?

RFC compliance

Fair enough, maybe, but links to actual incompatibility incidents would be helpful.

Linux features

This is a strong claim.

Requires vendor-specific tools

The "dramatic increase of support costs" is unsubstantiated FUD.

Poor user support

FUD again, presumes that only LKML people are can provide support, as opposed to the TOE hardware or system vendors.

Short term kernel maintenance

This is not an argument against TOE technology per se, just against Linux's design limitations that impede merging it.

Long term user support

Long term kernel maintenance

Both these presume that such abandoned hardware would somehow be broken. But TOE is easily disabled, so at worst old machines would support running without it.

Eliminates global system view

Reasonable point (though not particularly different from losing insight due to clever hard disk interfaces or graphics cards or whatever).

Fche 17:15, 19 August 2009 (UTC)