| “select *” run faster | “select *” run slower as need to combine data |
| Seek slower if not index. Traverse to block row 3, read all column until R3C5 (5 steps) | Seek faster. Traverse to block column 5, read until R3C5 (3 steps) |
| Aggregation slow, whole row of data is read out to memory before aggregate | Aggregate faster, only use column that needed. |
| faster write, jus append | harder to write; normally has a buffer write before commit into disc. |
| Saving empty value into data block (1 block = 1 row). | Did not save anything to data block because each cell in column has row key. |
| hard to compress data | easier to compress data as data in single column are more alike |
| RDMS has row size limit of 8kb, due to page size. | NoSQL can store bigger row but for big file/data, advise to store in hadoop and kept the file path here. |
| RDMS has number of column limit | NoSQL no column limit but you will face OOME |
Kim 2 ML 
Leave a comment