[openstack] support for modelling port binding attributes

[James Page]

OpenStack has the ability to support different types of ports for instances - by default these are typically fully virtualised at the hypervisor level and presented as a virtio-net port to the instance.

The datapath for this type of port always goes via the hypervisor kernel so is limited in terms of capacity and performance.

Its possible to use fully offloaded ports for instances with some types of network card (such as Mellanox ConnectX 5 or later) where a VF is mapped directly to the instance using PCI passthrough and steered via the hypervisor using openvswitch - bypassing the hypervisor kernel and maximising throughput and performance.

This requires a slightly different set of commands to create the port with a specific binding profile, and then use it with an instance as its created.

It would be great if this could be modelled in Juju in some way - its possible to mix and match different types of ports depending on the performance requirements. Hardware offloaded ports are a finite resource on a cloud (whereas virtio based ones are not/are but not by the NIC hardware).

OpenStack documentation: https://docs.openstack.org/neutron/queens/admin/config-ovs-offload.html#validate-open-vswitch-hardware-offloading

[John A Meinel (jameinel)]

Sounds interesting, and we can do some discussion of it as to use cases, how to model it going forward. Though in the short term I don't see quick wins towards this.

[Canonical Juju QA Bot (juju-qa-bot)]

This Medium-priority bug has not been updated in 60 days, so we're marking it Low importance. If you believe this is incorrect, please update the importance.

[Haw Loeung (hloeung)]

Any chance we could consider this for an upcoming cycle? It would help us reduce manual provisioning steps for VMs we provision for high network throughput (and disk I/O in some cases).