We are testing with clustered columnstore indexes on SQL 2014 CTP2 in our data warehouse to see how its performance stacks up compared to traditional rowstore type tables. However, we ran into a problem with one of the commonly used queries right away as it performed much worse compared to a traditional rowstore table. I was also able to replicate the issue using AdventureWorksDW2012 restored onto SQL 2014 CTP2 and the database set to compatibility mode 120.
the sample query is a star join to dimDate and dimProduct on FactProductInventory and filtering on a single day with the filter applied to dimDate.FullDateAlternateKey
In short, we noticed that when using a traditional row store, a star join can effectively use the seek predicate (or predicate pushdown in a case the dim column is not part of some index) while performing a clustered index seek on the fact table. However, when using clustered columnstore, it first performs a full clustered scan on the columnstore fact table (in row mode) and then performs a date filter in separate steps, which leads to much slower performance in my real-world scenario.
obviously the performance is fine if the date filter is directly applied onto the fact table instead of dimDate, but I don't believe that's a reasonable workaround especially for dimensions other than date