Re: CFT: new trunk(4)

From: Peter Jeremy <peterjeremy_at_optushome.com.au>
Date: Thu, 12 Apr 2007 17:11:29 +1000
On 2007-Apr-11 23:27:18 +0200, Arne Schwabe <schwabe_at_uni-paderborn.de> wrote:
>
>>Trunking is a way of combining multiple physical interfaces to increase
>>the bandwidth.  Trunking multiple VLANs on a single interface doesn't
>>make sense to me.
>>  
>Cisco calls this Trunk (multiple vlans over one physical connection 
>(with dot1q)). Combining multiple physical links is called channel. 
>Maybe that is were the confusion comes from.

Mea cupla.  I knew that - maybe I should do a better job of getting
the brain into gear before responding.  I'll justify my incorrect
terminology by claiming that it seemed consistent with the usage
implied by the original poster.

>>At least some of the proprietary protocols
>>are fairly dumb and just round-robin MAC addresses between the
>>physical links rather than dynamically sharing traffic across the
>>available links.  The former means that if most or all of your traffic
>>is for a single MAC address, you don't actually gain anything by
>>having multiple physical links.
>
>I have seen things break if you do real round robin,

To clarify, my reference to "round-robin" may have been unclear.  The
equipment I've seen will assign a MAC address to a physical port as
part of the MAC learning process.  All traffic to that MAC address is
then forwarded via that port.

> some pxe boot stuff 
>and other embedded tcp/ip stack which are intended for local network use 
>only don't like if packets are out of order,

I believe that the Ethernet standard requires in-order delivery.  This
makes real dynamic traffic sharing non-trivial.  The round-robin port
assignment ensures in-order delivery and will probably achieve
reasonable load balancing if traffic is distributed across a number
of MAC addresses.

-- 
Peter Jeremy

Received on Thu Apr 12 2007 - 05:11:32 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:08 UTC