Re: NIC on morollan



On 27/10/2004, at 2:02 AM, Linda Winkler wrote:

> where can I get more details on Intel chipset/driver issues and 
> traffic loss?
> lw

Linda,

I don't know if  the issue is real or not, rather just an idea I've had 
and for which I'm looking for more evidence. We have some machines here 
with different brand gigabit NICS which showed different multicast loss 
patterns in the dast beacon display, even though they were on the same 
subnet. Additional board swapping further supported the proposition 
that something was amiss with the e1000 driver/chip.

The problem is exemplified by the loss pattern shown by 
morollan.ncsa.uiuc.edu in the display at http://beacon.dast.nlanr.net/, 
namely the morollan row shows maybe a quarter of the entries as having 
a partial loss (about 50% at the moment). Machines without the problem 
show either 0 loss or NA, e.g. the machine ag-bridge.vislab.uq.edu.au 
previously used its onboard e1000 driver/chip and showed similar 
display to morollan. Its onboard NIC has since been disabled in favour 
of a plugin NIC which uses tg3 driver.

According to the theory, other machines in the dast beacon display 
which have the problem include ntgrd.depaul.edu, 
ag-capture.vrac.iastate.edu, ri-av.informatics.indiana.edu, 
mtsugrid1.mtsu.edu, agaudio.ndsu.nodak.edu, jan.envision.purdue.edu, 
magic.itap.purdue.edu, q.uicomp.uic.edu, keyshadow3-huba.kcpa.uiuc.edu, 
yendi.ncsa.uiuc.edu, beacon.net.unc.edu, limestone.uoregon.edu, 
ibgp.gw.utexas.edu, beaker.canet4.net, thorn.es.net

I've found only 1 possible counter-example, that of 
ag0-display.imb.uq.edu.au. It is using an onboard e1000 driver/chip and 
while it initally showed the problem quite clearly, it looks quite good 
at the moment - occasionally some losses of 1 or 2.

chris

>
> At 10:52 AM 10/26/2004, Christoph Willing wrote:
>> Mitch,
>>
>> I see that you're the contact for morollan.ncsa.uiuc.edu - can you 
>> confirm/deny whether its NIC uses the Intel e1000 chip? Its output in 
>> the central_loss page shows a pattern of losses which we think can be 
>> attributed to this driver/chipset.
>>
>> chris
>>
>>
>> Christoph Willing                        Ph: +61 7 3365 8350
>> QPSF Access Grid Manager
>> University of Queensland
>
>
>
Christoph Willing                        Ph: +61 7 3365 8350
QPSF Access Grid Manager
University of Queensland



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index