Kafka 101 - how does it achieve high throughput?