DA Morgan wrote:
> Joel Garry wrote:
>
>> Keith wrote:
>>
>>
>>> How about this.... create a Pro/C program where you can pass the number
>>> of instances you want to run -- say 100. Then the Pro/C program can
>>> fork 100 times, connect to Oracle , and execute the procedure with
>>> appropriate values for fromX and toY.
>>
>>
>>
>> Wouldn't that not be any better, since Oracle is going to have to latch
>> the system tables to serialize each table creation? Perhaps even much
>> worse, as the latch queues grow and grow and... I think this would just
>> be a grotesque denial of service attack, if it can even create that
>> many processes. Then you'll deadlock as the underlying system tables
>> need to extend (even LMT's given 500000 tables).
>>
>> Since some apps have tens of thousands of tables I don't think the
>> requirement is _inherently_ ridiculous, but I don't think it can be
>> sped up, either. And it might be ridiculous anyways, not enough detail
>> to know.
>>
>> jg
>> --
>> @home.com is bogus.
>> http://inquirer.stanford.edu/2005/jstaffor/woz.html
>
>
> I'm sticking with inherently ridiculous until there is a better
> explanation. ;-)
>
> In fact I'll up the ante. I say it is intrinsically, explicitly and
> patently ridiculous.
FWIW, I have seen a design recently requiring 1.5M stored procedures and
1M tables. (that's million,not milli :-)
The SAP schema suddenly looks boring....
Cheers
Serge
--
Serge Rielau
DB2 SQL Compiler Development
IBM Toronto Lab
Received on Fri Dec 09 2005 - 01:02:22 CST
