When I read MySQL vs PostgeSQL benchmarks my question was always: "How close is that to real life workloads. My real life workloads." I don’t want to argue about how close TPC-C test simulates real scenarios for a majority of MySQL installations. The only thing this test tells is how well MySQL runs TPC-C Test. That test suits MySQL very well: database doesn’t have any variable length text fields; total database size is small. But if you try to run forum or portal software, that is there MySQL has major installed base, you will see completely different picture.
At the Page2RSS.com we have experienced very poor MySQL performance. Workload is essentially one ~2Gb table with dozen of fields. Couple of fields is variable length text fields with up to 32Kb of data. There are 1-2 inserts per second and up to 10 read requests per second. We used to have MySQL 5.0.24, innodb, no transactions. Now we have PostgreSQL 8.2.0. Server is PC-class Linux box: P4 2.8GHz, 1Gb memory, SATA disks - noting special at all.
Graph below is a Google's Time spent downloading a page (in milliseconds).
That sharp drop in the middle of December is when we switched to the PostgreSQL.
