Re: help with horrible network failures

[Nivedita Singhvi <niv@us.ibm.com>]

Rob Gardner wrote:

> My current configuration has:
> # CONFIG_E100_NAPI is not set
> 
> Isn't that the same as it being disabled? Or should I change it to:
> CONFIG_E100_NAPI=n

That means NAPI is off, but from what Jon said earlier,
it's ignoring the compile option.

> I'm afraid that isn't easy to accomplish. I am working with another 
> researcher in a faraway land, and so I do not have direct control over 
> their machine. Changing their nic could take a while.

That's ok - it wasn't critical.

>> Sorry, ethtool can't be used to disable/enable NAPI. Ignore
>> this idiot, neuron misfire...

> So,,, what's the conclusion?

Er, that was just saying that ethtool couldn't be used
to alter this - you had to recompile the kernel in order to
disable/enable NAPI in the e100. And it's unclear whether
you are actually disabling it since it is always on. Not
sure what the maintainers intended, there, but claimed to
be fixed very recently..

However, I think you have two problems here - the
ASSERTION is one of them - but not the fatal problem - the
bad address in wakeup_common is the fatal error.
It would help if you could run some debug code (?),
unless someone on the Xen team or on this list has seen
that one before and knows what's going on..

>> Your sysctl settings would be helpful, too..

None of the following is very relevant to your
oops, but will help in your testing..

Could you turn this on? (this might affect the
assertion - but it's just playing with the race
window, really)..

> net.ipv4.tcp_low_latency = 0
echo 1 > net.ipv4.tcp_low_latency


BTW, if you are running httperf, you would probably
benefit from increasing your socket buffer sizes from
the default.
> net.ipv4.tcp_rmem = 4096    87380    174760
4096 109568 109568

> net.ipv4.tcp_wmem = 4096    16384    131072
4096 109568 109568

> net.ipv4.tcp_mem = 12288    16384    24576
122880 163840 245760


> net.core.optmem_max = 10240
102400

thanks,
Nivedita




-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click