LAGing backbone links

Wed Apr 6 22:17:01 UTC 2011

On Tue, Apr 05, 2011 at 08:05:59PM +0100, Nick Hilliard wrote:
> Some older equipment will unequally prefer certain links over others, 
> depending on the number of members in the LAG.  I.e. a 2-member LAG might 
> load balance equally under ideal conditions, but a 3-member LAG might 
> naturally load balance 2:2:1.

Even newer gear does that. TurboIron 24X for example. Some Force10
switch model(s) as well, no clue how old though.

LAGs have one big advantage over ECMP: with gear implementing
"minimum-links" feature, you can make sure your LAG bandwidth doesn't
fall below a certain capacity before being removed from IGP topology
so you can make sure redundant (full!) capacity elsewhere can automatically
kick in.

With ECMP traffic engineering and capacity/redundancy planning
becomes... "interesting". Aside of all the operational problems
regarding troubleshooting (traceroutes/mtr do love such ECMP hells) and
operational consequences of having a lot of adjacencies and links.

For all those reasons, I usually prefer LAGs (with LACP) above ECMP, even
when that means "more bugs" (vendors tend to not properly test all their
features on LAGs too).

Best regards,
Daniel

-- 
CLUE-RIPE -- Jabber: dr at cluenet.de -- dr at IRCnet -- PGP: 0xA85C8AA0