Automatic vertical-database design

An automatic physical-layout designer for a database-management system determines the database's physical layout from a set of training queries, the database's logical design, and a parameter k that indicates how many storage nodes can be lost without losing access to any of the data. The designer lays the database out as a column store such that the stored columns constitute redundant projections on the system's different storage nodes. It repeatedly identifies a projection, whose addition to the design will result in the greatest performance improvement for the training queries. In doing so, it takes into account the different compression formats to which the different projections lend themselves. When a projection has been identified as one to be added, it is added on one node, and k projections having the same columns are added to other nodes. The designer continues thus adding projections until a space budget has been reached.