Well just saying something is slow is a little hand wavy. I understand that writing a database in Clojure/Java is not great (hello Cassandra, Hadoop, Hbase, ...

Well just saying something is slow is a little hand wavy. I understand that writing a database in Clojure/Java is not great (hello Cassandra, Hadoop, Hbase, ....) but in my Clojure code I found STM rather useful for sharing the same reference and it was always enough fast for the use cases I was running into. I guess Datomic might be slow, I was more referring to the idea of using a transactor to manage the concurrent writes as opposed to the model the was proposed in the video. Splitting up your database to different parts makes sense to me.