[bess] ECMP, equal/unequal Load-Balancing, and https://datatracker.ietf.org/doc/html/draft-ietf-bess-evpn-unequal-lb-21

["Jeffrey (Zhaohui) Zhang" <[e-mail]>]

Hi Neeraj,

ECMP is mentioned many times in the draft. It seems to be used somewhat inconsistently.

My understanding is that ECMP, spelled as Equal Cost Multi-Path, is about the underlay paths to egress MHES PEs. If an ingress PE does not have equal cost paths to those PEs, then load-balancing will not be used at all by default - the closest PE is always used. This is to optimize the traffic delivery in the core network.

Of course, one may choose to do load-balancing even w/o ECMP, for better utilization of MHES ACs.

The above is regardless of whether equal or non-equal distribution of load to different PEs is used or not.

I suppose the document is mainly about equal or unequal (weighted) load-balancing, which can be applied regardless of whether ECMP exists or not (depending on the operator's choice).

If my understanding is correct, the document should minimize the use of ECMP term. Perhaps just mention it once or twice - either that ECMP is the prerequisite, or load-balancing (equal or unequal) can be used whether ECMP exists or not.

Some examples of inconsistent uses:

   Once consistency of 'Value-Units' is validated, ingress PE SHOULD use
   the 'Value-Weight' received from each egress PE to compute a relative
   (normalized) weight for each egress PE, per ES, and then use this
   relative weight to compute a weighted path-list to be used for load
   balancing, as opposed to using an ECMP path-list for load balancing
   across the egress PE paths.  Egress PE Weight and resulting weighted
   path-list computation at ingress PEs is a local matter.

The above paragraph says, "a weighted path-list ... as opposed to an ECMP path-list". It implies that ECMP refers to "equal" load balancing.

   While incorporating link bandwidth into the DF election process
   provides optimal BUM traffic distribution across the ES links, it
   also implies that DF elections are re-adjusted on link failures or
   bandwidth changes.  If the operator does not wish to have this level
   of churn in their DF election, then they should not advertise the BW
   capability.  Not advertising BW capability may result in less than
   optimal BUM traffic distribution while still retaining the ability to
   allow an ingress PE to do weighted ECMP for its unicast traffic to a
   set of egress PEs.

The "weighted ECMP" above implies ECMP is about the ECMP in the underlay.

   In an EVPN IRB (Integrated Routing and Bridging) overlay network as
   described in [RFC9135], with a CE multi-homed via a EVPN all-active
   multi-homing, bridged and routed traffic from ingress PEs can be
   equally load balanced (ECMPed) across the multi-homing egress PEs:

   *  ECMP Load-balancing for bridged unicast traffic is enabled via
      aliasing and mass-withdraw procedures detailed in [RFC7432].

   *  ECMP Load-balancing for routed unicast traffic is enabled via
      existing L3 ECMP mechanisms.

The above seems to refer to equal load-balancing.

Jeffrey

Juniper Business Use Only