from Hacker News

Real-Time Databases Explained: Why Meteor, Rethink, Parse and Firebase Don’t Scale

by DivineTraube on 7/16/17, 6:31 PM with 2 comments

• by dkarapetyan on 7/16/17, 6:57 PM

  If the problem you are solving doesn't map to being embarrassingly parallel then you will indeed have problems. This is not restricted to just "real-time", or more technically streaming DBs. Coordination has high overhead no matter what DB you are using.

  Also, I don't see how their design avoids hot keyspace issues. Traditionally this is the main problem with partitioned designs because you move the problem one level up to properly designing the key that will be used to distribute the load across partitions.

• by voidmain on 7/16/17, 10:21 PM

  This approach is "scalable" in the sense of partitionable, but it's still O(N^2)! Needing 9x the servers for 3x the users is not really "scalable" in the colloquial sense.

  An often more attractive approach is to partition over the keys of each database index.