Heikki Siltala wrote:
>
> I have been playing around with tools like Spotlight and what the
> database keeps waiting is the disks. It has been the same ever since the
> database was created over 4 years ago. Adding DBWR slaves, relocating
> redos and archived redos, tuning log buffer etc have been tried but the
> bottleneck is still the IO system which consists of "multiple SCSI disks
> put into a multiple disk arrays and connected to the server via various
> SCSI channels". What else could be expected from this kind of a mess.
>
> The dual 440 MHz 64-bit RISC processors of the current server never run
> at 100 %. What we get is typically 3 % to 9 % CPU load per a data
> warehouse operation. On read operation the database waits for single and
> multiple block reads and on write operation it waits for redo logs. In a
> new system I want to make sure that redo writing wont be the bottleneck
> again. I have been thinking of placing them on two 2 GB solid state
> disks, and give two 15 krpm SCSI disks just for the archived logs.
>
> Yes, what we really need is an opportunity to talk with a storage
> engineer. Since we are in Finland I'm not expecting much. So far all we
> have got is a) guys that don't ask a word about the application and just
> try to sell their own brand/solution and b) guys that don't ask a word
> about the applicaton and just wanting to use the same solution that they
> have been implemented to another customer or that they have been found
> using a Google search. :-)
>
> --
> Heikki
What else do you expect?
Processor<->memory is a thousand times faster than disks.
Tune (your statements) to do less IO, or learn to live with it.
And Spotlight (Quest?) has a history of wrong calculations;
a version would help. Still very much the tune by hit ratio
school, I'd say
--
Regards,
Frank van Bortel
Top-posting is one way to shut me up...
Received on Wed Nov 30 2005 - 05:38:21 CST
