Re: Understanding tsearch2 performance

On 07/14/10 14:31, Oleg Bartunov wrote:
> Something is not good with statistics,  91 est. vs 8449 actually returned.

I don't think the statistics difference is significant - it's actually
using the index so it's ok. And I've run vacuum analyze just before
starting the query.

> Returning 8449 rows could be quite long.

You are right, I didn't test this. Issuing a query which returns a
smaller result set is much faster.

But, offtopic, why would returning 8500 records, each around 100 bytes
long so around 8.5 MB, over local unix sockets, be so slow? The machine
in question has a sustained memory bendwidth of nearly 10 GB/s. Does
PostgreSQL spend much time marshalling the data through the socket stream?