I can't speak to the numbers postgresql can or cannot do but the numbers above sound very very doable. If you can get a hold of greg smith's postgresql high performance, I always liked his method of tuning buffers and checkpoints using the background writer stats. All of which can help with the IO load and caching.
good luck!
On Thu, Feb 12, 2015 at 4:55 PM, Kevin Grittner <kgrittn@ymail.com> wrote:
> Can you give any hint on the configuration and on the underlying > hardware?
Well, this particular web site has millions of hits per day (running up to about 20 queries per hit) from thousands of concurrent web users, while accepting logical replication from thousands of OLTP users via logical replication, so you probably don't need equivalent hardware. If I recall correctly it is running 32 cores with 512GB RAM running two PostgreSQL clusters, each multiple TB, and each having a RAID 5 array of 40 drives, plus separate controllers and RAID for OS and WAL.
For server configuration, see these Wiki pages for the general tuning techniques used:
The best course to solve your problem would probably be to review those and see what might apply, and if you still have a problem pick a specific slow-running query and use the process described here:
