Processor contention(?) and network bandwidth on AMD

[Joshua Baker-LePain]

On Mon, 29 Apr 2002 at 5:07pm, Robert G. Brown wrote

> Try this:
> 
>   Do your netpipe transmitter on an unloaded host, a host at load 1 and
> at load 2.
>   Do your netpipe receiver on an unloaded host, a host at load 1 and one
> at load 2.
> 
> Fill in the matrix -- load 0 to load 0, load 0 to load 1, etc.
> 
> I found (in similar tests done years ago) that a TRANSMITTER could be
> loaded to 2 (per cpu) with only a small degradation of throughput, but
> loading a RECEIVER would drop throughput dramatically, by as much as
> 50%.

I will indeed do these tests and post a followup.  One thing, though, that 
I don't know that I made clear.  I understand and accept that a higher 
load is going to affect bandwidth.  What I'm pointing out (and wondering 
why) is that the effect is *far* greater on dual AMD based systems than it 
is on, e.g., the dual PIII systems I have.

As an aside, At Mark's suggestion, I reniced the ksoftirqds on a S2466 
based system, and saw vast improvement.  For no load, 2 matlabs, and 2 
matlabs+2SETIs I saw (with the cursed 2.4.9-31 RH kernel):

ksoftirqds reniced to 0:

11463.5 KB/real sec
10637   KB/real sec
9585.39 KB/real sec

And reniced to -19:

11481.8 KB/real sec
10632.7 KB/real sec
9347.31 KB/real sec

FWIW.

-- 
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University