The key inputs for cluster sizing: throughput (MB/s producer), replication factor (× RF for replica traffic), retention (GB), and latency SLA. A single broker can typically handle 100–300 MB/s. For 1 GB/s with RF=3: 3 GB/s replica traffic → ~10–15 brokers. Storage per broker = (daily_throughput × retention_days × RF) / broker_count. CPU is rarely the bottleneck — I/O and network are. Partition count: aim for 50–100 MB/s per partition.
Key configuration
# Rule of thumb:
# Partitions = max(producers, consumers, throughput/50MB)
# Brokers = max(RF, partitions/100, storage_required/broker_capacity)