[KLUG Members] Network speed question.

[adam at morrison-ind.com]

> It doesn't matter what OS your using the answer would be the same.

I'm not certain this would be true.  The IP stacks of different OSs 
differ quite
a bit, or used to, in what produced the best performance.  I'd suspect the
prevelance of web services has caused allot of them to clean it up however.

> It's not a Linux verse Windows verse whatever OS issue, it's a 
> protocol / communications issue.
> P.S. If you really want to get super technical, Yes 127.0.0.1 would 
> be faster cause there is no hardware to talk to, it's a kernel

I doubt that if you have an IP address of a.b.c.d on eth0 and you send 
a packet
from the local box to a.b.c.d that it ever touches the hardware.  I'd suspect
routing would simply memcpy (or whatever the kernel equivalent is) into that
interfaces stack.  The difference I see is that if you have iptables rules on
one interface and not another that would CERTAINLY effect performance.

> network interface, verse 10.23.32.122 which is attached to an actual 
> nic across the pci/isa bus. So, in truth, 127.0.0.1 would be like 2-3 
> millionths of a second faster, but who cares? <grin>

If it does actually touch hardware the difference would be bigger,  the packet
would cause bus activity and interrupts.  Those are very expensive in the
perspective of system performance.  CPU/Core stuff is fast,  but even on the
newest boxes when you touch the bus your dropping orders of magnitude in
throughput - and the bottom half of device drivers are invasively preemptive -
that is, they invoke a context swtich.  Context switches devour cycles and,
obviously, suspend application processing.