Re: low-MHz server
Douglas A. Tutty wrote:
On Mon, Feb 04, 2008 at 11:18:50AM +0800, Bob wrote:
I'd get a modern ish server and underclock it, that way you'll be able
to get more RAM and bigger hard drives, the Athlon XP was fairly easy to
get down to 300 MHz with the FSB still @ 133, I never tried lower but I
don't see why not, for comedy value see if you can get the CPU clocking
lower than the RAM.
I've been looking at all this. There are many "desktop" or
"workstation" boxes around but they don't have lots of room. This will
be important in the future. 100 MHz boxes aren't going to have a
magical come-back so whatever box I get will have to last well itself,
but also be able to be extended. Consider that 8 years ago, having 9 GB
was huge, which a workstation would have while a server would have 12 of
those 9 GB drives. Now, you need a 9 GB drive just to have room for
stuff and still be able to compile patches.
I want a box into which I can plunk new hard drives without the BIOS
complaining. I think this means SCSI which is more likely to mean a
server; as long as one can change the interface on the back-plane to
hook up to a faster scsi card to match the new drives. Oye.
Right now there's nothing on eBay like this. There are Proliant 2500
and 5000 workstations and IBM PC 300s, but nothing with lots of upgrade
I wouldn't worry too much about the actual case, you can always change
it, my current server is in a super cheap
which has no hard drive bays at all but it has 8 5.25" bays that I can
put drive bays in.
I would _like_ (but not _need) the box to have PCI so that I could add,
e.g. a USB card if it didn't have it built-in. Ditto faster SCSI (or
maybe SATA if it could connect, ditto multi-serial ports).
Go PCI (33 MHz) and PATA (100 MHz ??) while they're cheap (SATA has a
link frequency of 1.5GHz)
I was also wondering, re RF/EMF shielding, if a rack-mount server in a
half-height rack with front and back doors may be a good way to go.
With the doors closed, there's a lot fewer openings large enough for the
EMF to get out.
If your wife is genuinely sensitive (and I don't discount it, who'd've
thunk we had a compass in our nose) I'd try to work out what frequencies
effect her and use an oscilloscope to work out what emits RF at the
undesirable frequencies, if she's not really sensitive but it's
psychosomatic the placebo effect of you stalking the house with an
oscilloscope and ebaying the microwave, hair drier, all your florescent
lights & a couple of items she's suspected in the past will probably
cure her, as long as you take it seriously and "scan" all new purchases
I'm not sure, at the hardware level, how underclocking works. Does it
slow down all electrical activity or does it just divide the clock down.
I know that this would reduce the bulk of the EMF frequency, but the
clock could still be going full-tilt. 300 MHz is still too fast. I
want to stay under 200 and closer to 100 MHz.
It works the other way round, my Front Side Bus runs at 133 MHz and my
Athlon XP has a multiplier of 18 so it has an operating infrequency of
2394 MHz, I think the lowest multiplier on Athlon CPUs is 5 so if you
can get the FSB down to 33 MHz you'd have an operating infrequency of
165 MHz, quite a lot of this extreme over or underclocking isn't
available from BIOS and would have to be changed by a utility after
booting, I only overclock under 'dows for gaming so I don't know what
utility you'd use but I know the linux nvclock tool can underclock my
old nVidia GPS and GRAM to under 100MHz. One advantage of this is you
can have a cron job run in the middle of the night, when you're asleep
on the other side of the house, set to bring the FSB back to 133 and the
multiplier back to 18, before performing any CPU heavy tasks and
returning to low RF mode before morning.
According to http://fab51.com modern 65nm AM2 CPUs also go down to 5X
Looking on the OpenBSD platforms page, the hppa lists lots of
workstations but not servers under the PA-7100 and PA-7150 processors
(the rest run too fast). HP's documentation on old servers is
I know that the problem is that there was a narrow time-slot when
servers were in the 100-200 MHz range which had the capabilities I need
today and into the future, which are still available and supported on
I'll keep my eye on eBay, my ear on misc@, and wait to hear from J.C.
Roberts to see what he has in his lab.
My pleasure, good luck.