On Thu, 10 Apr 2008, PFC wrote:
... Lots of useful advice ...
- If you often query rows with the same gene_ref, consider using
CLUSTER to physically group those rows on disk. This way you can get all rows
with the same gene_ref in 1 seek instead of 2000. Clustered tables also make
Bitmap scan happy.
In my opinion this is the one that will make the most difference. You will
need to run:
CLUSTER gene_prediction_view USING gene_prediction_view_gene_ref_key;
after you insert significant amounts of data into the table. This
re-orders the table according to the index, but new data is always written
out of order, so after adding lots more data the table will need to be
re-clustered again.
- Switch to a RAID10 (4 times the IOs per second, however zero gain if
you're single-threaded, but massive gain when concurrent)
Greg Stark has a patch in the pipeline that will change this, for bitmap
index scans, by using fadvise(), so a single thread can utilise multiple
discs in a RAID array.
Matthew
--
Prolog doesn't have enough parentheses. -- Computer Science Lecturer
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
