Broadcom NetXtremeII Packet Loss/PE1950

[Matt Saladna]

Ben,
	Thanks for the response.  Surprisingly, none.  The servers hover around
a 0.5 - 1.0 load average with ~350 processes and push on average about 3
- 4 Mbps each.  One of the servers being tested with the BCM5708 chipset
was recently hooked up a few weeks back with nearly zero load (pushing
80 Kbps).  On the switch side everything is peachy.  No errors are being
logged on the Cisco switch nor is link status being list.  I'm at a loss
as to what the cause may be.

As an aside, I doubt it has any importance with the problem, but Trac's
FastCGI handler exhibited excessive timeouts when trying to communicate.
 I traced it down to mutex timeouts from a busted processor code in the
5130s.  A microcode update from either September or August (can't recall
which) fixed the mutex timeouts and Trac with FastCGI hums along now.

- Matt

Ben Ruset wrote:
> What sort of load are you putting on the box?
> 
> I have a fleet of 15 PE1950's all running VMWare Server. When the box
> gets under load we'll see the Broadcom nic's lose link with the switch,
> taking all of our VM's offline for a few seconds.
> 
> I'm in the midst of swapping dual port E1000's into all of the 1950's we
> have. As soon as we did that on two of our machines, the problem went away.
> 
> 
> Matt Saladna wrote:
>> Hi,
>>     Has anyone noticed sporadic packet loss with the BCM5708 NICs on
>> Linux?
>>  I have been attempting to track down the root cause, but everything is
>> turning up empty.