So yesterday I was lifting big, heavy things and then resting between sets by powering up an assortment of
CPU and motherboard combinations and measuring their power consumption. Typical activity
here at the treehouse.
I picked up a
P3 International Kill-A-Watt Power Meter
for measuring the power consumption at the outlet and, apart from the motherboard, CPU, and
memory, the only thing hooked up is a DVD-rom with a Slackware Linux boot disk in it. For a
power supply, I used a Macron MPT-400, not because it's especially good, quiet, or one that I
would be likely to use in any of the new server or workstation machines--it's just been on my
test bed since 400 watts was big for a power supply, so why waste one of the particularly
good and quiet power supplies I've bought since just for testing purposes.
Oh, wait--this is not a P3 International Kill-A-Watt Power Meter I'm picking up here
I let the systems boot up and then logged in, but that's about it. Not a very stringent test, but a quick and
dirty one. Maybe too dirty.
|Athlon x2 3800
(Energy efficient version)
|Core Duo T2300e
||Asus N4L-VM DH
|1.1GHz Pentium III
|233MHz Pentium II
(plus Matrox Mystique 4Meg PCI Video Card)
Which is really a lot tighter clustering than I would have expected. Even the big EATX Supermicro
server board with 2Gig of ECC registered memory and 64-bit PCI-X slots doesn't require significantly
more power with two PIII CPUs than the single Core Duo laptop CPU on a little, frill-free MATX
motherboard that's theoretically intended to be a low-power platform. I ended up leaving the
Supermicro dual-CPU system running for a few hours while I was working on other things and when
I came back, it was then showing 45 watts of power consumption. I didn't go back and test what
happened over similar spans of time with the other boards.
So my grand hopes for slashing the power consumption across the board on the servers out here
might be a little dotted if not actually dashed--but it does look like "upgrading" lowlier machines
like the DNS servers from Pentium-II platforms to even a mobile-on-desktop solution would increase
rather than decrease power consumption, and the DNS servers just don't need that much oomph.
Unfortunately, I didn't have any of the hungrier system boards I've been using set up for easy testing
on hand--I've got some dual Athlon MP boards and dual P-IV Xeon systems in service, but no extras
just standing by already assembled with CPUs and ready for this test. For that matter, I have
plenty of dual PIII Xeon systems and a couple of quad PIII Xeon systems that I could check out, too,
but my enthusiasm hasn't grown to that level yet.
Because, at that point, I noticed an undesirable rattling and scraping sound coming from my own
webserver (the one you're reading this off of) which, upon inspection, was suffering from a CPU
cooling fan about to go out. I think it's an all-too-serious case of "if it's not broke, don't fix it" thinking
that has resulted in the main webserver still being a dual 266MHz Pentium II system with 384 meg of
memory. State of the art stuff...about ten years ago.
So I figure I'll turn in early and get up in the middle of the night to fix it. Except after I replace the
CPU fan, it won't reboot--the power comes on and then immediately shuts off again. Strangely, it
will power up as long as I disconnect one of the drives, so my first thought was that there was a
short between the SCA-to-68-pin adapter on that particular hard drive.
But replacing that didn't help, nor did replacing the drive, nor the cable. After trying many combinations
of swapped-out components, it finally turned out that replacing the power supply would allow it to
boot up *with* all the drives connected whereas the old power supply would shut down after half
a second unless one of the drives was disconnected.
So, now being able to reboot the webserver (which had been running for nearly a year since the
last time I'd rebooted it--on a power supply that, according to the date on it, was from 1998), it
let me know that one of the drives in the operating system array had failed. (Yes, it's still using an
array of 1-gig narrow SCSI drives for the operating system.) Fortunately, I had five drives of the
same model on hand to replace it with (even though I'd given away almost all the small SCSI
drives in the last few years). Unfortunately, even though they were the same model, they were
slightly newer drives and were 8 megabytes smaller than the older drives of that model. (Seagate
likes to do this a lot. This apparently serves no purpose except to prevent you from using newer
drives to replace the older drives in RAID arrays.)
Since the drives were 8 meg smaller, the RAID controller refused to use them and I finally found a
drive in my spares cupboard that wasn't smaller--a 4.3 gig drive. So the smallest replacement drive
that I had which the controller would accept was bigger than the enire array.
At which point I rebooted the system and found that the partition table of the data drive array
had been overwritten with all zeros. So I ended up repartitioning, reformatting, and restoring
that array. And now we're back.
So even if the newer and greater hardware I was testing yesterday isn't going to save a lot of
power over my ten-year-old current webserver, maybe this is just nature's way of hinting that
there may be other reasons to upgrade that box.
So how was your morning?