9:55 not sure how updating a value in row oriented db if the value is bigger than the previous value, will be slower than in column oriented db. Column oriented db will also require other values in the column to be shifted wouldn't it. Also how is a bulk/batch update faster in column oriented db. Each record will still be updated one by one and will require other fields in the column to be moved around for each update.
I can sees that Columnar databases are theoretically slower at OLTP, but with the speed of modern hardware and the fact that the two obvious OLTP processes are small by their nature (i.e. inserting a row from a POS system, or deleting a row, something I prefer not to do -- just mark the row as deleted with a flag), surely this theoretical time penalty is not a real practical limitation on using a Columnar database for OLTP processes? It would be interesting to see a set of real world timings that compared real OLTP business processes conducted on a row based and a columnar based database.
