Re: [Banana] diff (banana load distribution, next-hop selection of multipath)

[Margaret Cullen <[e-mail]>]

> On 2/4/2017 9:49 AM, Margaret Cullen wrote:
>> Hi Joe,
>> ,,,
>> There are all sorts of devices that are routers AND include other sorts of traffic shaping or loadsharing.  There are even some routers deployed today that can share a single flow across multiple access links, recombining the traffic in a provider network.   However, there is no standard that I am aware of for doing this, so both ends need to be provided by a single vendor.
> 
> OK, so this would be about picking a standard for coordinating the
> endpoints, tunneling with sequence numbers, and (maybe) the "ECMP"
> sharing alg.

Picking/developing one (or more) standard(s), yes.  A few different solutions have been proposed, and we quickly reviewed them during the BOF.  The two major classes of solution under discussion on the BANANA list have been bonded tunnels (with packet resequencing and flow control at the bonded tunnel endpoints), or something MPTCP Proxy-based (some with various extensions for handling UDP/QUIC traffic).

Because there are significant trade-offs between these two types of solutions that make them applicable for different scenarios, it is possible that we might end up deciding that it makes sense to standardize one of each class.  Or, maybe one of them can be improved to become the preferred solution for all scenarios.  I don’t think we know how that will come out yet.

> I'm just trying to scope the problem. Right now, it looks a lot like it
> might also include trying to use multiple heterogeneous links without a
> cooperating party upstream where the traffic merges, which is quite a
> different problem.

No, that has been tried in the past with very poor results.

> IMO, it'd be useful to be clear about that scope.

Thanks!  I will try to make this clearer in the problem statement.

Margaret