On Wed, 12 Aug 2015, Juliusz Chroboczek wrote:
- interference avoidance: do we want to balance over any two paths (in
which case no signalling is needed), over non-interfering paths (in
which case the Babel-Z extension provides all you need), over any
wired paths (again, Babel-Z is enough), or over fully disjoint and
non-interfering paths (in which case I'll need to extend Babel-Z in an
incompatible manner or design a new extension)?
To avoid complication, I would be fine with just saying "if two directly
wired connections of the same speed is established between two routers, do
ECMP" and don't do ECMP in any other case unless manually configured (I
have no idea how much the last part makes things complicated).
Basically, trying to get ECMP to work over media with different speed and
different characteristics sounds very complicated.
Yeah, but if you do it naively, performance will get worse, not better.
(The IS-IS community has it easy, they just implement things and don't
care if they make things worse.)
"caricaturing".
Folks, if anyone has done a biblio search for ECMP, I'll be grateful for
pointers. If not, I'll take an afternoon of my copious free time and do
it myself. (But don't hold your breath, I've got my hands full with
analysing the Battlemesh results and getting shncpd into shape, and I
have a serious teaching load that starts in early September).
Generally ECMP is done when IGP metric ends up identical or when there are
static routes with the same "metric" pointing to the same place. At least
in the reality I've been exposed to.
It can also be done over BGP, but this is less common.
I don't know exactly what you're looking for, ECMP isn't that complicated:
https://en.wikipedia.org/wiki/Equal-cost_multi-path_routing
So basically:
A--B
| |
C--D-E
If metrics of ABDE and ACDE is the same, then program FIB to split L4
flows between these two paths by means of for instance hashing.
--
Mikael Abrahamsson email: swm...@swm.pp.se
_______________________________________________
homenet mailing list
homenet@ietf.org
https://www.ietf.org/mailman/listinfo/homenet
