2010/4/9 Scott Marlowe <scott.marlowe@xxxxxxxxx>: > On Fri, Apr 9, 2010 at 10:30 AM, Merlin Moncure <mmoncure@xxxxxxxxx> wrote: >> On Fri, Apr 9, 2010 at 12:03 PM, Greg Smith <greg@xxxxxxxxxxxxxxx> wrote: >>> The main problem with this configuration is that work_mem is set to an >>> unsafe value--1.6GB. With potentially 400 connections and about 2GB of RAM >>> free after starting the server, work_mem='4MB' is as large as you can safely >>> set this. >> >> if you need more work_mem for this or that and also need to serve a >> lot of connections, you can always set it locally (1.6GB is still too >> high though -- maybe 64mb if you need to do a big sort or something >> like that). >> >> Another path to take is to install pgbouncer, which at 400 connections >> is worth considering -- but only if your client stack doesn't use >> certain features that require a private database session. zabbix will >> _probably_ work because it is db portable software (still should check >> however). > > Also remember you can set it by user or by db, depending on your > needs. I had a server that had a reporting db and an app db. The app > db was set to 1 or 2 Meg work_mem, and the reporting db that had only > one or two threads ever run at once was set to 128Meg. Worked > perfectly for what we needed. > Thanks for all Your advices. I will set up new parameters on Monday morning and see how it perform. Greetings for all PostgreSQL Team -- Krzysztof Kardas -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
