1. In the case of 100 brokers:
- Your cluster is fully distributed and can handle tremendous volumes, even using commodity hardware.
- Zookeeper may be under pressure because of many open connections, so you need to increase the Zookeeper instance performance
- Cluster management is a full time job (make suer no broker acts weirdly)
- It is not recommended to have a replication factor of 4 or more, and this would incur network communications within the brokers. Leave it as 3
- Scale horizontally only when a bottleneck is reached (network, disk i/o, cpu, ram)
댓글 없음:
댓글 쓰기