Re: Pentium M and Oracle Performance

[EMAIL PROTECTED] (Daniel Gustafsson) writes:

> > And no, I can't explain those results - with a single
> > 60 GB IDE disk, the Notebook should be much slower.
> > (It's even faster with an external 200MB USB2 drive, btw.)
> 
> Most IDE disks shipped today, especially in notebooks, have their
> write cache enabled. I don't know how your RAID setup looks like (do
> you have write caches with battery backup?), but it may be unfair to
> compare a caching notebook disk with a RAID setup.

All sorts from different manufacturers. With >= 64 MB of battery-backed
controller-cache, too.

> It is more difficult to explain the USB drive performance. I don't
> know how fast USB2 is, but it is likely that the 200MB disk (IDE?)
> also have the write cache enabled.

I still think the the Oracle-OS interaction is the culprit.
Maybe some notebook specific hack to let the internal drive
stay powered down for longer periods. Though if the API calls
work as designed, Oracle should certainly get write-through
semantics.

On the other hand, it might be just the drives. At >= 40MB/sec
sequential read and write throughput, a single IDE drive
is often faster that the ill configured and outdated RAID
hardware we get to see in x86 servers. (On the notebook
drive I measured just 13MB/sec seq write and 16MB/sec read,
though. I don't have values for the 200GB USB drive).

Regards,
Chris