"Shridhar Daithankar" <shridhar_daithankar(at)persistent(dot)co(dot)in> writes:
> On 29 Jan 2003 at 13:34, Francisco J Reyes wrote:
>> Just checked my values and I have 128MB for vacuum and sort. Is that
> Absolutely. Especially sort mem. If you are allocatingg 128MB sort mem for each
> connection, you will soon be dry with memory. I would say 8MB to 16MB sort
> memory depending upon number of conenctions you want to entertain at a time.
> Just remember it is per connection.
Worse than that: it's per sort step (or hash step). It's not at all
unlikely that an N-table query would be running N-1 sorts or hashes in
parallel; plus maybe more to implement aggregation, grouping, ORDER BY,
etc. So unless all your queries are very simple, you'd better figure
on some multiple of sort_mem per connection.
>> It seems I only have 2MB of WAL buffer.. I have large daily loads of data.
>> Will a number above 10MB help? ie 16MB
> 16MB is good.
16MB is probably vast overkill. Are you able to measure any change in
performance at all from increasing wal_buffers? I can't see a reason
that more than a dozen or so wal buffers would be useful. Almost
certainly, those megabytes would be more effectively used as
general-purpose kernel disk cache.
regards, tom lane
In response to
pgsql-general by date
|Next:||From: pginfo||Date: 2003-01-30 15:47:57|
|Subject: Re: URGENT: referential integrity problem|
|Previous:||From: Marc G. Fournier||Date: 2003-01-30 15:32:53|
|Subject: Re: donations?|