Forum OpenACS Development: Re: OpenACS performance issues

22: Re: OpenACS performance issues (response to 21)

Posted by Matthew Coupe on 03/06/08 02:59 PM

On my dev box I changed the Master Template back to Zen and I was still getting some slow request times on the permission checks.

429 ms 0or1row dbqd.acs-tcl.tcl.acs-permissions-procs.permission::permission_p_not_cached.select_permission_p: 0or1row nsdb0

select 1 from dual
where exists
( select 1
from acs_object_party_privilege_map ppm
where ppm.object_id = '3551196' and ppm.party_id = '14554' and ppm.privilege = 'read' )

23: Re: OpenACS performance issues (response to 22)

Posted by Gustaf Neumann on 03/06/08 04:13 PM

zen is no performance bottleneck, it won't affect the speed of permissions at all.

we have on our learn system more than 10 times the number of entries in the permission tables, and much better speed (most time below 1ms, with up to 2000 concurrently active users; but we have also a different hardware).

The indices you have are correct, these are the indices of the standard openacs installation.

We have on our productive environment two indices more:
Try to add these:

create index acs_privilege_descendant_map_privilege_idx on acs_privilege_descendant_map (privilege);
create index acs_object_context_index_object_id_idx on acs_object_context_index (object_id);

after creating the indices, do a "VACUUM <tablename>" on both tables.

measure again. It is a while, since i have added these indices. They might be an artefact of our local modifications. if they don't not help, remove the indices again.

24: Re: OpenACS performance issues (response to 23)

Posted by Matthew Coupe on 03/06/08 05:01 PM

zen is no performance bottleneck, it won't affect the speed of permissions at all.

Sorry, I wasn't very clear. I was benchmarking against Zen to see if there are any extra expensive queries in my custom Zen template. Zen is a little faster but that is to be expected as our custom template is a little more complex.

I added the indices but there was little effect and some queries were still taking 240ms+.

Dave mentioned increasing Shared_Buffers from 32MB - I tried this but Postgresql wouldn't restart. What sort of value should I be going for here? Could this be the issue?

26: Re: OpenACS performance issues (response to 24)

Posted by Gustaf Neumann on 03/06/08 06:19 PM

Seems twice as fast, but not really fast.
Have you set shmmax as indicated above? shmmax must be larger than the shared buffers, otherwise postgres won't start. note, that if you have multiple postgres servers running, you need memory for each server. What is the message, when postgres refuses to start?