Grasshopper said "f you are looking at aggregating a single distinct metric. If you need to have many metrics (measures) in a row, then RDBMS is faster so calculations can be performed across the row in the Kernel. Row Calculations is where this technology fails."
>5 years ago we moved from SQL Server to Infobright. While there were several compromises that we had to take into consideration moving to a columnar database, calculations across rows was in no way one of them. I literally have hundreds of reports and most of them have at least 5 metrics. these reports typically run in < 4 seconds. Largest fact table approx 175 million rows.
Compromises that we had to take into consideration was that Insert Update and Delete were horribly long. We overcame this by utilizing drop and rebuild table or partitioning (In our BI Software not in the database).
One Huge advantage of the Columnar database was/is that we really do not have to be as selective of what we include in the table. if the table is 100 columns wide it is irrelevant because there is no IO overhead on columns not selected in the query. The other huge advantage of Columns is that I do 0 (Zero, Zilch, Nada) maintenance/performance tuning on the database. nice for a small shop!
We are, in fact, going to move back to SQL Server 2016 if the Column Store Indexes are as improved as the Marketing folks claim.