Can you please post to netdev? Posting networking related issues to
linux kernel alone is considered rude. Posting them to netdev only
is acceptable.
> Ihar 'Philips' Filipau wrote:
>
> >Werner Almesberger wrote:
> > Ihar 'Philips' Filipau wrote:
> >
> >| | | Modern NPUs generally do this.
> >
> >
> > Unfortunately, they don't - they run *some* code, but that
> > is rarely a Linux kernel, or a substantial part of it.
> >
>
> Embedded CPU we are using is based MIPS, and has a lot of specialized
> instructions.
> It makes not that much sense to run kernel (especially Linux) on CPU
> which is optimized for handling of network packets. (And has actually
> several co-processors to help in this task).
The coprocessors are useful, but that has nothing to do with the value
of the NPU. You can add those within a general processor system.
I am also in the camp that to be really useful these things need to run
a real OS - Linux.
> How much sense it makes to run general purpose OS (optimized for PCs
> and servers) on devices which can make only couple of functions? (and no
> MMU btw)
>
> It is a whole idea behind this kind of CPUs - to do a few of
> functions - but to do them good.
>
> If you will start stretching CPUs like this to fit Linux kernel - it
> will generally just increase price. Probably there are some markets
> which can afford this.
>
Actually i believe it will lower the prices.I am waiting for intel to get
hyperthreading right - then we'll see these things disapear.
The only thing useful about NPUs is their ability to management the
discrepency between memory latency and CPU speeds. Trust me i used to
be in the same camp as you.If you note, a lot of these things appeared
around the height of the .com days. VCs were looking for something
new and exciting.
> Remeber - "Small is beatiful" (c) - and linux kernel far from it.
> Our routing code which handles two GE interfaces (actually not pure
> GE, but up to 2.5GB) fits into 3k. 3k of code - and that's it. not 650kb
> of bzip compressed bloat. And it handles two interfaces, handles fast
> data path from siblign interfaces, handles up to 1E6 routes. 3k of code.
> not 650k of bzip.
If all you wanted was to do L3 - why not just buy a $5 chip that can do this
for a lot more interfaces? Why sweat over optimizing L3 routing in a
3K space?
to nit: Its no longer about routing or bridging, friend. Thats like getting
fries at mcdonalds.
cheers,
jamal
|