I have been trying to find out more about Voldemort but it doesn't seem to be used outside of LinkedIn. Has anyone actually used it for a real system?
Gilt is also using Voldemort. Check the InfoQ talk.
I was the responsible of a project related to turism industry that was using Voldemort as the storage of night/price availability of hotel rooms. At the time I was in the project, there were more than 8K hotels. For each hotel there were (not all combinations):
-1 years of nights (from today in advance)
-7/8 different rates (advance purchase, etc.)
-5/6 combinations of occupancy (2 adults, 1 adult).
-4 or 5 taxes.
The values where continuously updated (for price changes, etc.).
Stored in a 4 Voldemort clusters (big fat machines in a double CPD).
From my experience there, Voldemort has a slow learning curve in terms of how to properly setup it for optimal performance (a lot of adjustments in the configuration, etc.), but once it as set, the performance was blazing fast. I don't remember exactly the numbers, but our system was receiving more than 40 req/s. For each request there were a mean of 30 hotels, which mean that Voldemort was returning 2400 items per sec when the keys were available (1200 hotel rooms + 1200 hotel taxes).
As I say, the trick here is able to setup properly voldemort, and design proper access patterns (a good data model), and I guess this is mandatory for all database systems (sql or nonsql): you will not gain the best performance (even you will get a poor one) if your data model is bad.
Yes - A friend at a major, well-known Internet company (not LinkedIn) is using Voldemort. The project is still under development.
Initially, Cassandra was developed by facebook. They are not using it much, one of the reason why most of the people don't choose it!
Also technically, Voldemort is phenomenal in read speeds. So a user store/a preference store would essentially need much of read speed than write speed.
Sometimes Cassandra can be a over-killer when you just need a key/value pair store.