Downstream Attention Summary
new Column Store Index option for creating indexes in Compass

PM Ticket Summary
Add a new type of an index, columnar index, that substantially improves performance of analytic queries.

Motivation
Document-based collections with B-Tree indexes ensure fast writes and point or selective range reads. They are efficient transactional and search scenarios that process entire documents. On the other hand, analytic queries usually reference a handful of fields and perform well against column-oriented formats. In this case, fields are highly-compressed and can be loaded selectively providing dramatic savings in I/O. Other techniques, such as skip-scans or operations on compressed data can push performance even further.

Allowing users to supplement their collections with columnar indexes is a seamless way of adding support for analytic workloads to their systems trading off storage and write throughput. In addition, it can improve performance not only for a known upfront fixed workload but also for ad hoc queries.