For long time we've employed Oracle database. As soon as the amount of data and workload had increased, it has become obvious that query and table structure optimization won't be enough. At the very least, we had to enable partitioning which is an additional paid option in Oracle. Ideally, we should've migrated to some sort of MPP solution which costs crazy amounts of money at Oracle (grid, RAC, Exadata).<br /> <br /> We've looked at different options and decided to take PostgreSQL as the basis for our platform, set up a single powerful server alongside the Oracle and gradually migrate, and afterwards transition to the MPP solution Greenplum.<br /> <br /> The job took 4 years. It was constantly complicated by the increase in workload and statistics, and the fact that there was a separate product present that utilizes the database heavily.<br /> <br /> During the migration we've implemented many custom solutions such as:<br /> - an on-line replication of data Oracle -> PostgreSQL;<br /> - a system for database patching and application synchronization and start-up;<br /> - a system for import of large volumes of statistical data;<br /> - a monitoring solution, as well as a tool for PostgreSQL similar to Oracle Enterprise Manager;<br /> - a migration to the Pentaho platform for this database;<br /> - numerous other systems which implemented specifics of interaction with jobs, partitions, stand-bys and backup solutions.<br /> <br /> At the present moment we have finished migration to PostgreSQL. Certain statistics is being stored at Hadoop. Some of the disks had to be made un-RAID in order to fit the ever growing volumes of data. Question of migration to Greenplum remains to be discussed. Total team size - up to 30 engineers, including 6 database developers. At the moment the database size is about 3 Tb, not considering the statistical data stored in Hadoop.