Hi all, One of the problems I been thinking about for quite a while is how to improve the efficacy of cassandra's gossip, especially when dealing with large clusters (greater than 1,000 nodes). By reducing gossip overhead we can implement new features that require data to be disseminated cluster-wide and we can support those features to thousands of nodes. To that end, I've done much research and thinking, and have opened up CASSANDRA-12345 as a parent ticket for this work. I'd like to invite the community to discuss the ideas, the approach, and hopefully dig into the implementations that I've posted in the child tickets.
Thanks, -Jason