Re: extremly low memory usage

From: John A Meinel <john(at)arbash-meinel(dot)com>
To: Jeremiah Jahn <jeremiah(at)cs(dot)earlham(dot)edu>
Cc: postgres performance <pgsql-performance(at)postgresql(dot)org>
Subject: Re: extremly low memory usage
Date: 2005-08-18 02:21:04
Message-ID: 4303F090.40602@arbash-meinel.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

Jeremiah Jahn wrote:
> I just put together a system with 6GB of ram on a 14 disk raid 10 array.
> When I run my usual big painful queries, I get very little to know
> memory usage. My production box (raid 5 4GB ram) hovers at 3.9GB used
> most of the time. the new devel box sits at around 250MB.
>
> I've switched to an 8.0 system on the new devel box, but the .conf
> really didn't change. Index usage is the same. Something seems wrong and
> I'm not sure why.
>

How big is your actual database on disk? And how much of it is actually
touched by your queries?

It seems that your tough queries might only be exercising a portion of
the database. If you really want to make memory usage increase try
something like:
find . -type f -print0 | xargs -0 cat >/dev/null
Which should read all the files. After doing that, does the memory usage
increase?

>
> any thoughts,
> -jj-
>
>
> shared_buffers = 32768 # min 16, at least max_connections*2, 8KB each
> work_mem = 2097151 # min 64, size in KB

This seems awfully high. 2GB Per sort? This might actually be flushing
some of your ram, since it would get allocated and filled, and then
freed when finished. Remember, depending on what you are doing, this
amount can get allocated more than once per query.

> maintenance_work_mem = 819200 # min 1024, size in KB
> max_fsm_pages = 80000 # min max_fsm_relations*16, 6 bytes each
> checkpoint_segments = 30 # in logfile segments, min 1, 16MB each
> effective_cache_size = 3600000 <-----this is a little out of control, but would it have any real effect?

It should just tell the planner that it is more likely to have buffers
in cache, so index scans are slightly cheaper than they would otherwise be.

> random_page_cost = 2 # units are one sequential page fetch cost
> log_min_duration_statement = 10000 # -1 is disabled, in milliseconds.
> lc_messages = 'C' # locale for system error message strings
> lc_monetary = 'C' # locale for monetary formatting
> lc_numeric = 'C' # locale for number formatting
> lc_time = 'C' # locale for time formatting
>

John
=:->

In response to

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message Gavin Sherry 2005-08-18 02:30:24 Re: limit number of concurrent callers to a stored proc?
Previous Message Jeremiah Jahn 2005-08-18 02:11:48 extremly low memory usage