svn commit: r292379 - in head/sys: netinet netinet6

[Kubilay Kocak]

On 18/12/2015 9:55 AM, Steven Hartland wrote:
> On 17/12/2015 19:20, Gleb Smirnoff wrote:
>>    Steven,
>>
>> another feasible solution for the design described in the 156226
>> would be to run STP on the switches, and if_bridge(4) instead of
>> if_lagg(4) on FreeBSD, also with STP enabled. Would work perfectly.
>>
>> Of course, if switches are dumb and cheap, and can't do STP,
>> then a tiny bpf-writer is the right solution.
>>
>> P.S. When I was running network in my university dormitory, we
>> used a lot of cheap solutions, and a lot of dirty workarounds,
>> but none of the latter made its way to FreeBSD kernel. You can
>> also ask Eugene Grosbein, he also has huge experience of living
>> on not so pleasant workarounds, but not pushing them agrressively
>> into the kernel.
>>
> Last time I heard STP is a bad word in networking, so I'm sure they
> network team
> would have me crucified for even suggesting it and start shouting MLAG
> for the
> rest of the day ;-)
> 
>     Regards
>     Steve

FWIW, during my testing at $lastjob, STP-enabled switches didn't provide
optimial failback times, even with portfast enabled.

This was for two FreeBSD machines providing HA network management
services for multiple backends configured with dual-port bge/em NIC's
configured with lagg(4) (bge0/em0, bge1,em1) in failover mode (no
aggregation), with multiple carp(4) interfaces for public service IP's
on top.