Why Pentium 512

Cost is one major factor, considering notebook drives tend to be around $1 per GB still and the proposed fileserver will have at least a TB mirrored. Then there's the 40-44 pin adapters dangling in air behind the drives, something I'd like to avoid for reliability's sake.

I'd expect the notebook drives to be the most significant bottleneck, that I can underclock-undervolt the CPU and mainboard and still have over 30MB-s with desktop drives. It won't be necessary to underclock it but can't hurt in achieving long lifespan either, the system can remain viable for at least 10 years on GbE and a PCI based PATA & SATA card(s). Whatever I come up with for AC power conditioning will probably cost more than the entire system (sans drive costs). Plus, if I keep power consumption below a certain threshold, I have plenty of parts for building a redundant power board. Beyond that threshold, it becomes yet another expense, and same situation with some UPS I already have. I could justify the expenses if necessary, if I didn't have any fileservers yet but I do, so... It becomes more a matter of a creative exercise during a routine pre-planned obsolescence of one of my older fileservers. They run fine but I'd rather replace while they work rather than after one had failed, especially when it can be done at my leasure instead of an immediate necessity. That just allows playing around a bit before it's done.

Essentially I'll be leveraging the most cost effective modern technologies and ignoring those with a high cost:benefit ratio. I can underclock to lengthen service intervals (except the inevitable drive failures over several years use) and promote longest motherboard lifespan. A light duty filesever just doesn't need much in the way of CPU performance and the software raid may be an issue but on a similar existing box with a Celeron 500, it never reaches 100% CPU utilization more than momentarily, fractions of a second with averages (during transfers) well below 75% (I forget the exact percentage at the moment).

If higher CPU performance had significant gain I could just put a higher performance platform to use but I expect the biggest bottleneck to be the 32bit-33MHz PCI bus as it was the case with post-1GHz Tualatin fileservers and it would raise costs by an order of magnitude to overcome that PCI limitation with a newer PCI Express or ported GbE platform & complimentary CPU. I'll have to benchmark it, maybe it won't perform well enough underclocked but it's hard to speculate as too few people have benchmarked C3 in these kinds of uses. I'd seen some filesystem benchmarks that make C3 look fine but not with software raid. So I'll underclock then compare and see...