Project Apollo is a new feature in CTP3 of SQL Server Denali that enables a new columnstore index that offers 10-100x performance improvements for a star join or similar query. Apollo brings together the in-memory columnstore technology (VertiPaq) that is used in PowerPivot and a new query execution paradigm called batch processing to provide impressive speed improvements for common data warehouse queries. Microsoft has claimed in test scenarios customers have experienced approximately 100x improvements in star join and similar queries.
VertiPaq makes columnstore indexes more efficient because it uses a different way of storing columns than traditional indexes, and it effectively compresses the data in the index. In a regular index, all indexed data from each row is kept together on a single page, and data in each column is spread across all pages in an index. In a columnstore index, the data from each column is kept together so each data page contains data only from a single column. In addition, the index data for each column is compressed, and since many columns often contain highly repetitive values, the compression ratio can be very high. This architecture reduces the number of pages in the index and, if you are selecting only a few columns, it also reduces the number of pages that need scanning (and therefore it is more likely that SQL Server will be able to keep them in memory).
Courtesy : Jamessera