High Capacity Single Table Performance Design Using Partitioning in Oracle or PostgreSQL.

摘要

This report describes the design rationale and resulting performance seen in a single server (non-clustered) database with an extremely large table containing in excess of 11 billion records. The original research and testing had been performed on a much smaller scale into the tens of millions of records around fall 2001, with an expectation that it would be near to the projected maximum number of records. In summer 2003, after those initial results were used to design a table still in use in production today, a similar table with a much larger data set was needed and the basic research was applied again to the new table. Initially, it was expected the number of records would peak close to half a billion; however, due to continuous and unexpected growth, the table now contains more than 11 billion records and is expected to grow more as the storage capacity is increased on the database server.