[sdnog] Common problem in sudan | Duplex mismatch

[Nishal Goburdhan]

On 28 Jan 2015, at 10:42, Tarig Yassin <tariq198487 at hotmail.com> wrote:
> 
> Hi Bernd 
> 
> > good point ... in general there is no magic - the problem is quite standard on 100 MBit interfaces when you don´t
> > define if you connect auto to auto or manual to manual - if you try to mix auto negotiate with manual duplex you
> > have for sure this problem.
> 
> how disappointing.. I didn't expect that, i thought it was vendor problem.

my memory is fuzzy here, but auto-negotiation was optional, and not mandatory for the ethernet (10bT) and fast ethernet (100TX) standards.
it *was* made mandatory for gigabit ethernet, so, you *should* expect issues with non gigabit ethernet equipment.
when i did run a network, we made it mandatory to hard-set 100FD (for 100 mb/s full-duplex) as needed;  ie.  do not auto-neg. 

it's technically out of specification for you to *force* a port to 1000FD, because you *should* let them auto-negotiate, but, in practice, i can tell you that this doesn't always work out the way you want it to.  today, at the JINX, for example, we still force 1000FD to gigabit port peers, as default.  

and why would you think your network problems are either sudan, or huawei specific?   we all have network problems don't make sense sometimes;  that's why we have community support    :-)


> > the check of duplex errors should be part of each link check after turnup - and as things are changing you should
> > also periodically check your interfaces on switches and routers ...
> 
> it seems this is the only way, but it can be through script, right?

yes, absolutely.  i'm sure scott will mention RANCID when he talks about logging and scripting, so i won't won't talk about it now.  but, also, if you have an NMS system (eg. cacti/observium) you can also use this to graph errors on the port  (it's built into both, iirc) and you should be watching those as well.  that's a Really Good thing to do.  graphs aren't just for bandwidth usage  :-)

--n.