We need to design a churn metric to retain the most valuable connections.
Proposition
We could start with a metric which is similar to upstreamyness metrics: count the number of messages first advertised by a peer, which were later proved to be valid (invalid messages must be ignored).
This will require changing the tx-submission to a protocol with reversed agencies like the object-diffusion protocol for Peras. Note that the object-diffusion protocol must be updated to enable client-side timely demotion.
Since this requires a protocol change, the new protocol must be negotiated in a new node-to-node version; connections that negotiated tx-submission can be at the bottom of the metric.
NOTE: this change requires updating the CIP#137.
We need to design a churn metric to retain the most valuable connections.
Proposition
We could start with a metric which is similar to
upstreamynessmetrics: count the number of messages first advertised by a peer, which were later proved to be valid (invalid messages must be ignored).This will require changing the
tx-submissionto a protocol with reversed agencies like theobject-diffusionprotocol forPeras. Note that theobject-diffusionprotocol must be updated to enable client-side timely demotion.Since this requires a protocol change, the new protocol must be negotiated in a new node-to-node version; connections that negotiated
tx-submissioncan be at the bottom of the metric.NOTE: this change requires updating the CIP#137.