In relation to my previous thread (about SW RAID vs. HW RAID on a P400), I was able to narrow down the filesystem speed and in general, our new system (running PG 9.1) is about 3x - 5x faster on the IO.
In looking at the query plans in more depth, it appears that PG 9.0 and 9.1 are both preferring to do hash joins, which seem to have a "linear" time and are slower than PG 8.4 doing an index scan.
For example, on PG 9.x:
Same query on PG 8.4:
Both of these servers have the same indexes, similar postgresql.conf, and almost identical data. However, the old server is doing some different planning than the new server. I've run analyze on both of these databases. Some relevant PG parameters:
max_connections = 150
shared_buffers = 6400MB (have tried as high as 20GB)
work_mem = 20MB (have tried as high as 100MB)
effective_io_concurrency = 6
fsync = on
synchronous_commit = off
wal_buffers = 16MB
checkpoint_segments = 30 (have tried 200 when I was loading the db)
random_page_cost = 2.5
effective_cache_size = 10240MB (have tried as high as 16GB)
If I disable the hashjoin, I get massive improvements on PG 9.x ... as fast (or faster) than our PG 8.4 instance.
--
Anthony Presley