This is interesting to me. What is the indexing strategy that is utilized?
To me the challenge is that the values making up a row are stored non-contiguously. The best case scenario is that a row of N fields requires N data accesses. This is assuming that you can index directly into an array or similar structure with constant access time.
Row-based storage means that you get all fields for "free"when accessing a single row, since the DB engine accesses a page at a time.
This is my understanding of the problem - I am very interested to learn what I'm missing if you have any good articles to share.
HBase + Phoenix is the closest I know of. Though, the cost of covered indexes is duplication. Phoenix also provides local/global indexes, giving an optimization choice b/w read heavy vs write heavy patterns
I think this is very accurate for first generation column stores, but some of the newer research systems are addressing this issue with technologies like behind-the-scenes indexing.