Kafka log writing

Question

Would anyone please explain to me how Kafka manages to keep the write throughput high? From my understanding the log is in append only mode hence there is no disk seek involved.

However,

1. For each <topic, partition>, two log files are maintained, one for the actual messages, the other is an index.
2. Each broker machine could hosts many <topic, partition>, and IIUC Kafka keeps their log files separated.

So it seems on one broker machine we still need to write to many output files, and then this "no disk seek" benefit seems to not be true?

Thanks!