Scaling HBASE *down* instead of *up*

I know that you can launch hbase in non-distributed mode. It's not clear from my reading whether it's practical to use it in production for a range from one node through a few up to a full-fledged cluster. The alternative would be to abstract out an interface and toss in something simpler in the small case(s). Ha开发者_运维技巧s anyone out there experience in deploying hbase on the low end of the scaling spectrum?

We started out in production with 4 nodes without problems, and have since scaled up to about 50 as our data grew.

I don't see any reason in theory why you couldn't run everything on a single node, so long as it had plenty of resources ( see can HBase , MapReduce and HDFS can work on a single machine having Hadoop installed and running on it? ), though I'm not aware of anyone who started that small.