> Relational databases scale. NoSQL databases scale. This is simply not true, if we're talking about horizontal scalability (which is probably the most relevan

> Relational databases scale. NoSQL databases scale. This is simply not true, if we're talking about horizontal scalability (which is probably the most relevant kind of scaling these days, because it's the most cost-effective). You can't add more nodes to an RDBMS, keep ACID, and get linear scalability. You just can't. From what I know of CouchDB, you could scale it linearly if you manage the partitioning of data yourself according to load. But if your data distribution changes you're a bit screwed. My impression though is that most people just use CouchDB on a single machine. For BigTable, on the other hand, it really is just a matter of adding or removing machines at any time and getting nearly linear scaling. The load is dynamically balanced between machines, so this is actually practical. Are there ways of scaling an RDBMS architecture? Yes, but they involve giving up ACID (like having read-only replicas, for example). So I'm not saying RDMBS's are bad, just that there is a real limitation here.