svn: stable/7/sys: conf dev/bce

[Maxim Sobolev]

Attila Nagy wrote:
> Maxim Sobolev wrote:
>> Stanislav Sedov wrote:
>>> On Mon, 30 Nov 2009 13:25:03 -0800
>>> Maxim Sobolev <sobomax at FreeBSD.org> mentioned:
>>>
>>>> Can we change the name then? We've seen the similar issue without 
>>>> the jumbo frames, which made me wonder.
>>>
>>> It's strange, because 'normal' frames are less than a page size in 
>>> length,
>>> so the issue I was experiencing should not be reproducable with jumbo 
>>> frames
>>> disabled.  Can you describe a bit more what kind of problem you're 
>>> experiencing?
>>
>> The traffic through interface abruptly halts randomly when under a 
>> load, without any signs of a problem. What is interesting is that the 
>> communication through unix domain sockets appears to be down too, 
>> since some scripts that connect to database not able to complete as 
>> well, so whatever this problem is it affect the whole network 
>> subsystem. We tried to replace hardware and it did not help. Only 
>> after we replaced bce(4) with IntelPro card (em) the problem has went 
>> away.
> Sorry, I've just noticed this thread and did not read the earlier posts. 
> I can see similar, although only (at least for now) on version 8 
> (running something around RC2), also with bce interfaces and "heavy" -in 
> terms of pps, this is a DNS server- network load.
> 
> Everything which involves network halts for some minutes, but the 
> machine is completely usable from the console. top shows some processes 
> (I don't have an output, but as far as I can remember they were only 
> that which did network IO) in the "keglim" state.
> Then something happens and the machine starts to work again.
> 
> Can this be the same, or is it completely different?

It really looks like the same issue. Those boxes are heavy-duty VoIP 
switches, so that they are handling high PPS as well in our case.

-Maxim