On this smaller test, the indexes are over the allowed memory size (I've got over 400.000 readings per sensor) so they are mostly written in disk. And on the big test, I had small indexes (< page_size) because I only had about 5-10 rows per table, thus it was 3000*8kb = 24mb which is lower than the allowed memory. btw which is the conf parameter that contains the previously read indexes ? I cannot do test this weekend because I do not have access to the machine but I will try on monday some tests. Thanks for your answers thought Selon Scott Marlowe <scott.marlowe@xxxxxxxxx>: > On Sat, Aug 23, 2008 at 1:35 PM, <tls.wydd@xxxxxxx> wrote: > > Actually, I've got another test system with only few sensors (thus few > tables) > > and it's working well (<10ms insert) with all the indexes. > > I know it's slowing down my performance but I need them to interogate the > big > > tables (each one can reach millions rows with time) really fast. > > It's quite likely that on the smaller system the indexes all fit into > memory and only require writes, while on the bigger system they are > too large and have to be read from disk first, then written out. > > A useful solution is to remove most of the indexes on the main server, > and set up a slony slave with the extra indexes on it to handle the > reporting queries. > > -- > Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-performance >
