Re: One particular large database application

On Mon, 22 Apr 2002, Curt Sampson wrote:

> On Sun, 21 Apr 2002, Francisco Reyes wrote:
> > May I suggest a different approach?
> > From what I understand this data may not change often.
> > How about instead of getting numerous cheap machines get only 2 or 3 good
> > machines with 2  15K RPM drives, 4GB of RAM and 1 IDE for the OS.
>
> It won't be as cost-effective. Price increases much faster than
> performance for an individual component. Take a look at CPUs for
> example:
>
>     CPU            Cost    $/GHz
>     2.4 GHz P4        $539    $224
>     2.0 GHz P4        $322    $161
>     1.9 GHz P4        $225    $118
>     1.8 GHz P4        $158    $88

The 1.8Ghz would be fine.

> Drives? 73 GB 10K RPM SCSI drives start at $399. I can't even find
> a 73 GB 15K RPM drive, but the 36 GB drives are $388 and up. For
> $400 I can buy four 7200 RPM IDE drives, and assuming I have them
> on separate controllers, I'm going to get much better aggregate
> throughput than I could ever get with a single SCSI drive. Not to
> mention that I end up with more than four times the storage space
> as well.

If you were to get 4 machines I would still think 2 machines with 15K rpm
would be better than 4 machines with 7200rpm IDE drives.

IDE's may have good thoughput, but their seek times can't compete with top
of the line SCSI.

> This is the great thing about distributed systems. The only trick
> is distributing your application as well. And that's often a pretty
> big trick, otherwise everybody would be doing it.

Exactly.. since it won't be easy for you to find the best distributions
then it may be worth getting better hardware. :-)