Set default effective_io_concurrency and random_page_cost to SSD values #613
Conversation
|
So far we've stuck with keeping the defaults equal to the upstream defaults and only overriding them as needed but I guess we don't have to. |
|
We've overridden this on all the nominatim machines and api DB servers, and I was about to do so for all the render servers, so I figure we might as well do it globally. Some defaults just aren't sensible on a server of reasonable capacity, since PostgreSQL aims for defaults that will work on a very small machine. |
|
Shouldn't this also remove the overrides that we already have in place then? |
Since all the drives with DBs on them are SSDs, set these GUCs to values more appropriate to a generic SSD. NVMe drives could have even higher concurrency, but this is a reasonable start.
pnorman
force-pushed
the
effective_io_concurrency
branch
from
5e50da8 to
4018bab
Compare
pnorman
changed the title
|
I've removed the overrides that were 256 or near it. Nominatim has per-server specific overrides, so I left those in place. cc @lonvia I also added random_page_cost since we were overwriting it on tile servers, all the api DB servers, and nominatim servers. I left the nominatim override in place, since that's been tuned to unusual values. |
Since all drives are SSDs in arrays with at least two drives, 256 is a reasonable default. NVMe drives could have this set even higher, but this is a reasonable start.