[RFE] Using multicast for image transfer

[Vladyslav Drok]

In huge environments network is bottleneck during simultaneous node provisioning. By starting group deployment over multicast will minimize network utilization. And will minimize resource utilization on image storing servers.

Baremetal nodes that are spawned with the same user image may be grouped. Since user image is equal on all nodes there is no need to send it by unicast directly to each node. Rather ironic should wait until all nodes in the group are ready to receive user image. Once it is done "user image" should be transfered to all nodes simultaneously via multicast.

[Vladyslav Drok (vdrok)]

Copy of whiteboard:

Gerrit topic: https://review.openstack.org/#q,topic:bp/multicast-provision,n,z

Addressed by: https://review.openstack.org/225570
    Using multicast for image transfer

[Jim Rollenhagen (jim-rollenhagen)]

Spec is abandoned, someone is welcome to pick this up

[Seamanhanshan (yaojun)]

I have some questions about this spec, if someone still work on it, maybe we could have a discussion.

Q1: Which multicast mode is considered? Any Source Multicast or Source Specific Multicast.
Q2: How many group address will be used? For example, One user image uses one group address(or channel).
Q3: How to control the receiver, not all the baremetal nodes can receive the user image.
Q4: Multicast isn't reliable, how about RMTP?