[j-nsp] Creating multiple LAGs Vs putting all links in a single LAG - Pros and Cons

[adamv0025 at netconsultings.com]

> NK NSP
> Sent: Friday, March 31, 2017 4:59 PM
> 
> Hi,
> 
> Just looking for some comments and/or thoughts on either approach. If you
> have 2 diverse fiber spans between 2 end-point IP devices with separate
> metro gears for each span and have created multiple 100G links on each
fiber
> span, would you rather create a single LAG bundle on both IP end-points
and
> put all 100Gs in single LAG or create 2 separate LAGs and put the links
from
> each span into its own LAG bundle (for e.g. ae0 for Span A and ae1 for
Span
> B).

Depends on how loss sensitive is the traffic being transported over the
setup. 
If you can afford 50-100ms loss of Nx100Gbps as a result of any member link
failure/removal and then also any member link restoration/addition, heck
sure one LAG at both ends and be done with it.

LAGs are tricky,
I'd suggest ECMP instead, loss is much less on link failover and restoration
with ECMP compared to LAG.
At minimum I'd suggest splitting the setup into two LAGs so that only part
of the inks is affected by the LAG weirdness in case you need to add a new
link in future or during failure.  

> Some of the motivations for single LAG is config simplification and less
> interface states for router to manage. 
Not a valid reason, unless you're planning to extend the model to a backbone
of 1000's of boxes, your router couldn't care less about the delta in number
of links and IGP states you'll have in this setup.

> For multiple LAGs one can argue
> better hashing and link utilization for all links.
Actually the same hashing algo is used for ECMP and LAG, so in theory there
should be no difference, but subject to test results   



adam
 


netconsultings.com
::carrier-class solutions for the telecommunications industry::