Forum OpenACS Q&A: Response to Development machine recommendations sought...

Posted by Don Baccus on
Oh, you need a real machine and a laptop both.  There's no question about it!

As I hinted in my response to Jade (we've drunk beer together, once over my laptop) I do work much of the time on my laptop.

But ... you still can't beat a machine with decent drives, ECC (for stability, my laptop has occassional mysterious Oracle problems that I don't see anywhere else, and I suspect the lack of error correction, and besides, I'm an old-fashioned old-fart).

Via's popular chipset for Athlon has tested out to be great, but I'm guessing Jerry's an SMP kinda guy.  There's something cool about dual processor machines, though my only one is my webserver (which is vastly underutilized but which I love anyway).

I suspect Jerry's into the cool factor, as I would be, even though he agreed with my assessment that switching to 15K RPM SCSI would be the most cost-effective thing to do (for me, that has high coolness factor since I'm running 7200 and 5400 RPM ATA and SCSI drives).

DDR and RDRAM have been shown to provide surprisingly little performance improvement over SDRAM alternatives for most applications.  You need to be doing a lot of memory shuffling and little computation.

Meanwhile Oracle is doing directly buffer-to-disk I/O bypassing the kernel buffer cache (at least Linux 2.4 has this feature) which diminishes the improvement in Oracle performance you can get with higher speed memory.  In PG, increasing the buffer cache to reasonable levels means you'll reduce shuffling, too.

So ... it's not clear in practice how much better DDR will perform than SDRAM for the kind of systems we deal with.

Better, obviously, but the "bang for buck" issue isn't clear.

I do agree with the comment that Intel's forcing a lot of mobo and memory rebuys.  I don't see it as being intentional, rather I see it as being a result of reactionary responses to the totally unexpected (by Intel and frankly everyone else) ability of AMD to not only challenge Intel on price/performance, but on raw performance.  About two years ago Intel was thrown into a tizzy by the Athlon and (worse, from their POV) the roadmap, and realized that their slowdown on incremental improvements in order to concentrate on a far-off "next generation" left them vulnerable.  So ... scramble, scramble, scramble and at the same time the rate of incompatible upgrades were accelerated.  Most consumers don't upgrade, after all ... they buy a new box.  And more importantly Intel had no choice from an engineering point of view.