On Wed, Aug 14, 2002 at 08:53:39AM +0200, Hermann Strassner wrote:
> > Thats true, sort of, but not really. PCI can BURST to roughly 1Gb/s,
>
> Standard PCI with 32 Bits and 33 MHz? I dont think so.
>
> > Given that a full duplex ethernet can eat up half of the real bus
> > bandwidth (if you have 2 it could dominate the bus),
>
> What sort of cache would use 400 Mb/s Network bandwidth constantly?
>
I had to read the original post 4 times before I figured it out. I think the
confusion is Mbits vs MBytes... The first post was talking 132 MBytes/sec,
and the respondant switched to 1Gbits/sec. I use a personal convention of
using MB for MBytes, and Mb for Mbits to try and avoid this confusion.
132 MB/s = 33MHz * 4Bytes = 33MHz * 32bits = 1056Mb/s ~= 1Gb/s
That's the burst rate. He then says the real bandwith after overheads etc is
closer to 400Mb/s. full duplex 100Mb/s = 200Mb/s, hence a saturated full
duples 100T NIC uses half of the real bus bandwidth of PCI.
> > I read on this list that someone
> > claimed that a dual processor didnt make much difference,
>
> > but most dual
> > processor MBs have 2 busses
>
> I havnt heard this, and i use more than one Dual PCs.
> Esp. not two PCI busses, of course they have two processor busses.
Because PCI is an unterminated bus that is "tuned" to minimise the affects
of reflections etc, it's length is critical can can't have too many devices.
I suspect anything with more than about 3~4 PCI slots actually has two
physicaly seperate PCI busses. With single processor, these are joined by a
PCI bridge, so they behave like a single PCI bus. With a dual processor, it
would make sense to make these two logicly seperate busses for each
processor.
Have a look at cat /proc/pci to see what you have got. Note that one of the
devices is usually a PCI bridge. Also note that the AGP slot is a seperate
logical "PCI" bus.
-- ---------------------------------------------------------------------- ABO: finger abo@minkirri.apana.org.au for more info, including pgp key ----------------------------------------------------------------------Received on Wed Aug 14 2002 - 04:52:06 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:09:37 MST