[lug] advice on a problem

Steve A Hart shart at colorado.edu
Sat Jul 31 11:52:40 MDT 2010


First thing I'm going to do is remove the olde PCI network card.  Dan is
quite correct that it is extremely old.  I only put it in to make sure
the much newer onboard NIC was not the problem.  Once I switch back,
I'll see if the overruns clear up.  My guess is that they won't since
the lockups occurred last time I had the onboard NIC active.

These are all great ideas.  I'm very much appreciative of the help everyone!

Steve

On 07/31/2010 10:24 AM, stimits at comcast.net wrote:
> There you go...overruns. I don't know how you could renice the network
> drivers to have higher priority, but that'd be an interesting way to
> test. What speed is the NIC and switch? *If this were in the old days*,
> and we were talking about transfers from a hard drive, I'd wonder if the
> DMA were on or not...with DMA off, copies of data were done in software
> through the cpu, and with DMA on, the transfers would be in hardware
> bypassing cpu...which would make a diff between large transfers
> consuming cpu like crazy, or no cpu activity at all to transfer data (in
> modern days, think of braindead winmodems and winprinters with hardware
> removed, versus real hardware).
>
> Does anyone here know if there is any kind of analogy in modern times
> for a NIC versus the old hard drive DMA (though I suppose a lot depends
> on the particular NIC being used)? Is it possible that the kernel is not
> servicing the NIC fast enough under load, due to a configuration? If the
> other systems work fine, yet this one has overruns using the same
> networking, it implies something along the lines of the kernel having
> too much network traffic on this one machine...perhaps it is getting
> more data, but more likely something is different on this machine.
>
> ...
> For my problem system:
> RX packets:243655 errors:0 dropped:0 overruns:104 frame:0
> TX packets:153291 errors:0 dropped:0 overruns:0 carrier:0
> ...

-- 
Steve Hart
Systems Administrator
Colorado Center for Astrodynamics Research
University of Colorado Boulder
shart at colorado.edu
(303)492-8109



More information about the LUG mailing list