[Teas] Modelling VPNs in Network Slicing and Receiver SLOs

[Adrian Farrel <[e-mail]>]

Hi,

The open question resulting from my long discussion with Luis concerns how
you model complex services and whether this requires the use of "receiver
SLOs".

We seemed to arrive at two possible motivations:

1. A sender is sending traffic volume N, but may break that into Na, Nb, and
Nc (Na+Nb+Nc <= N) to three different receivers, and may vary those three
volumes over time. It is possible for the sender to have three P2P
connectivity constructs with volume SLO set to N on each, but this may be
expensive and not completely helpful to the provider.

An option here is to have a "meta SLO" that spans multiple connectivity
constructs. Thus, we could express that the sum of the volume across all
three constructs will never be more than N, but the peak on any one
construct could also be N. (Presumably, the provider would police or provide
admission control.)

Another option is to use an A2A construct which leads to...

2. A customer wants to model a VPN service (such as L3VPN) in network
slicing. They do this using an A2A connectivity construct so that the
provider knows that there may be traffic from any SDP to any other SDP. The
sender SLOs tell the provider the volume from each sender, but not the
volume to each receiver.

It is possible that knowledge of the receiver AC capabilities is enough. It
may be that the "excess" traffic at the receiver is just discarded. Or it
may be that some form of receiver SLO is useful.

How do people who want to model VPNs in network slicing do this today?

Thanks,
Adrian