|
Horms wrote:
>
> On Fri, May 12, 2000 at 11:52:53AM -0700, Lars Marowsky-Bree wrote:
> > On 2000-05-12T20:37:33,
> > Nicolas Huillard <nhuillard@xxxxxx> said:
> >
> > > Can you define what is a "fast Intel box", on your point of view ?
> > > PIII 500, 64MB, IDE disks, 100Mbps NIC ?
> > > What is the bottleneck in such a config : CPU, memory, type of NIC ?
> >
> > My guess would be the PCI bus would be the first limit to hit, when
> > saturating
> > the ethernets...
>
> I would have thought that you would have to have quite a few NIC -
> even if they are gigiablt - to saturate the PCI bus. Then again
> hardware isn't my strong point.
>
> --
> Horms
>
Well, the theoretical bandwidth for a 32bit PCI bus is
132M*bytes*/second.
So when you are talking about 100M*bit*/second nics, this really isn't
an
issue at all. If you are doing gigabit nics then you (theoretically
again)
should be able to get full gigabit out of one, any more and you'll need
to
use a 64bit pci slot (which most gigabit cards now-a-days support for
2Gbit full duplex on 1 card). Although the most I've ever gotten out of
gigabit cards on Linux was around 650Mbit/second after a lot of
tweaking,
but that's a different issue involving the old 1500byte ether frame
size,
which is really inadequate for gigabit ether. I've heard some higher
numbers
from some guys who have tested gigabit nics that use 'jumbo frames' or
9k
ether frame sizes under linux. Only problem with that is, everything
connected to that nic needs to support 9k frames. :)
--
Dan Cox
Linux.com [VA Linux Systems]
Systems Administrator
|
