Re: 15,000 tables - next step

Jan Wieck schrieb:
> On 12/2/2005 6:01 PM, Michael Riess wrote:
>
>> Hi,
>>
>> thanks for your comments so far - I appreciate it. I'd like to narrow
>> down my problem a bit:
>>
>> As I said in the other thread, I estimate that only 20% of the 15,000
>> tables are accessed regularly. So I don't think that vacuuming or the
>> number of file handles is a problem. Have a look at this:
>
> What makes you think that? Have you at least tried to adjust your shared
> buffers, freespace map settings and background writer options to values
> that match your DB? How does increasing the kernel file desctriptor
> limit (try the current limit times 5 or 10) affect your performance?
>
>

Of course I tried to tune these settings. You should take into account
that the majority of the tables are rarely ever modified, therefore I
don't think that I need a gigantic freespace map. And the background
writer never complained.

Shared memory ... I currently use 1500 buffers for 50 connections, and
performance really suffered when I used 3000 buffers. The problem is
that it is a 1GB machine, and Apache + Tomcat need about 400MB.

But thanks for your suggestions! I guess that I'll have to find a way to
reduce the number of tables. Unfortunately my application needs them, so
I'll have to find a way to delete rarely used tables and create them on
the fly when they're accessed again. But this will really make my
application much more complex and error-prone, and I had hoped that the
database system could take care of that. I still think that a database
system's performance should not suffer from the mere presence of unused
tables.

Mike