We should base the piggyback transmission count (and protocol period length) off of the group size as in the SWIM paper. The default value was hard-coded to 5 which would only work for really small groups (10s of processes). Even so, some processes aren't getting membership updates, so there could be a bug in the SWIM dissemination. We should validate this.
In the meantime, piggyback tx count has been set much higher (50) to help ensure update propagation.
In GitLab by @shanedsnyder on Mar 12, 2021, 09:35
We should base the piggyback transmission count (and protocol period length) off of the group size as in the SWIM paper. The default value was hard-coded to 5 which would only work for really small groups (10s of processes). Even so, some processes aren't getting membership updates, so there could be a bug in the SWIM dissemination. We should validate this.
In the meantime, piggyback tx count has been set much higher (50) to help ensure update propagation.