Hi Ivan,
True, but then you need a mechanism to discover all the available
endpoints, whereas with single VTEP per host you can use existing
mechanisms (either learning from source IP address or next-hop
distribution with BGP-like protocols).
True.
I see few options:
- Signal different next hops in BGP for the application which runs across
- Treat next hop in BGP as prefix indicator not as host route. In
particular next hop resolves in IGP to a prefix which in turn could be
used as set of GRE destination addresses
- Signal such range explicitly. I think there will be a draft coming up
on just that soon.
True, but many existing DC switches that do IP-based load balancing
have no problems load-balancing on UDP 5-tuple. I am positive
there's equipment out there that can do load balancing on GRE keys
(or some other part of GRE header), I just haven't encountered it yet
(or realized it does that), so please fix my ignorance.
Indeed. In that case we have nothing to worry about. I think this thread
was just about the case where the above does not apply.
In fact if we all agree that devices can build hash looking deeper and
that those which can not are just a corner case that would be great.
Problem solved :)
And what is nice about IP encapsulation all GRE src/dst addresses
can be naturally aggregated so from IGP point of view they still
look like a single prefix.
... and you need a new endpoint prefix (or set-of-addresses)
discovery mechanism.
I don't need nothing on top on what I need today. I need to advertise
reachability to end point prefix anyway. The difference is that I
advertise for example /28 rather then /32.
Hi Stewart,
I think that this needs some serious study rather than assertion,
because it is not at all obvious what the LB properties of either
method will be in a DC.
The claim was much more general and my reply did not actually narrowed
it to DC env. I was more talking about WAN.
In the case of the use of multiple GRE addresses there are
operational and route scaling issues that also need to be considered.
I do not see any.
Smart networking OS of ingress and egress point can do such
encapsulation without any operational overhead. Please think in terms of
mGRE.
Route scaling issues do not apply. You do not need to inject more then
one prefix in any case.
Regards,
R.
