Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Fri, Mar 07, 2014 at 04:53:09PM +0100, drbd-user-request at lists.linbit.com wrote:
> Subject: Re: [DRBD-user] "PingAck timeout" in a dual active/passive configuration
Hi, we recently encountered strange link problems with round-robin
bonding on directly connected recent Intel GBit NICs (I350 if I remember
correctly). They would lose link every now and then, sometimes the
entire bond0 would stop working.
If that is the case in your setup, this would show up as "Link down"
messages in dmesg.
I suspected the default setting of EEE (Energy Efficient Ethernet) to be
at fault, but I can't test anymore as the system shipped to the customer
where nodes are connected using a switch, and the symptom never happened
again. That is one of the reasons why I suspect EEE - not many switches
support EEE, and EEE is automatically disabled unless both sides support
it.
The intel driver (at least in 5.0.5) supports EEE-Setting at module load
time using a module parameter:
modprobe igb EEE=0,0,0,0
(that would be for an Intel 4port card, one arg for each port).
With recent kernels, you might get ethtool support for dynamic EEE
querying/setting.
Regards
Matthias Ferdinand