ifconfig Carrier problem

Mike Heroux linux-thinkpad@www.bm-soft.com
Fri, 16 Jul 1999 16:09:55 -0500


I have a Thinkpad 770Z and use a 3COM Megahertz 10/100 LAN CardBus Model
3CCFE575BT.  I use it as part of a "Beowulf" cluster. 

I followed the instructions listed at the URL 

http://www.bm-soft.com/~bm/tp_pcmcia.html

and have been successfully using this system for several months.  However,
I do have some performance problems when I include the laptop as part of my
cluster computations (via LAM MPI).  All my results are correct but the
performance of the parallel job is very poor.  Looking at the output of
ifconfig, I see a very large number for the "Carrier" field on the laptop
and some overruns.  These values are zero on all other nodes.  I include
the output below.

I have already tried:

- Switching the ethernet cable.
- Configuring LAM with various nodes excluded.  My bad parallel timings
only occur when the laptop is one of the nodes used for the computation,
unless of course it is the only node being used.
- A different interrupt number.  The same results occur when I use
interrupt 9 (instead of 3 as listed below).

I am wondering if you have any guess as to what could be causing my
problem.  Any suggestions would be greatly appreciated.

Thanks in advance.

Mike

% ifconfig eth0

eth0      Link encap:Ethernet  HWaddr 00:50:04:5E:24:AF  
          inet addr:192.168.0.214  Bcast:192.168.0.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:749848 errors:0 dropped:0 overruns:1419 frame:0
          TX packets:804113 errors:0 dropped:0 overruns:0 carrier:793545
          collisions:0 txqueuelen:100 
          Interrupt:3 Base address:0x280 


Michael A. Heroux, Ph.D.
Applied and Numerical Mathematics Department
Massively Parallel Computing Research Lab
Sandia National Laboratories
ph: 320-845-7695
fx: 320-845-7846
mheroux@cs.sandia.gov