Even prior to the JSON type you always could have serialized the data for the sample to some kind of string JSON/XML/CSV and put it in the column. But I think

Even prior to the JSON type you always could have serialized the data for the sample to some kind of string JSON/XML/CSV and put it in the column. But I think the one thing NOSQL data stores e.g. HBase/Cassandra and to a lesser Degree MongoDB has taught me is to pay attention to the tradeoffs number of rows, size of index, etc...