The main difference between the two is how each one generates hash values. The random separator used the native JDK MD5 hash (because it was convenient for developers and standard for all JDKs). But since Cassandra really does not need a cryptographic hash, this function took much longer than necessary.
With the Murmur3 splitter, token hashing only does what Kassandra needs. Which consists in creating a marker that provides uniform distribution over the nodes. This leads to a 3–5-fold improvement in the hash performance of the token, which ultimately leads to a total increase of 10%, which Carlo mentioned above.
, DataStax , . , , , () . , Murmur3.