Well, the pair of greens I have (that I'm still using) one of the chassises did 
have a PS failure, replaced the PS and it booted up fine.  That server is 
backed up as are all of my servers.

The only time I've ever seen a PS failure that scotched the drive was an 
embedded device.  It was just a cheaply made PS

Cooling fan failures are common but generally take out the GPUs.  Most of the 
common desktop Dells and such you find around save a nickle
By using a combo giant case fan and ducting to cool a massive CPU heatsink.

Fan quality matters tremendously.  Fans in my Cisco network gear rarely fail 
I've only had 1 that did in a router.  There's 4 other fans in there I replaced 
it not because the router was overheating but because the log was filling up 
with entries bitching about the failed fan.  Fans in cheap crappy switches tend 
to fail often.

Ted


-----Original Message-----
From: Keith Lofstrom <[email protected]> 
Sent: Tuesday, December 9, 2025 6:07 PM
To: Ted Mittelstaedt <[email protected]>
Cc: 'Portland Linux/Unix Group' <[email protected]>
Subject: Re: [PLUG] Hardware advice: replacement internal desktop drive

On Mon, Dec 08, 2025 at 11:29:09PM -0800, Ted Mittelstaedt wrote:
> Only if it's a single drive.
> 
> I don't build SOHO servers anymore with single drives.  Nowadays I use 
> 2 drives and Ubuntu Server than during the setup configure BIOS to set 
> the disks as just a bunch of dumb drives, and then during OS 
> installation turn on drive mirroring.

Keep in mind that power supplies can fail as well, often killing what is 
connected to them. 

Usually failure is to zero volts output, but a particularly nasty form of 
failure is the output electrolytic capacitors going open circuit, often 
resulting in damaging power pulses well above the normal DC design average.  
Integrated ircuits can draw VASTLY more current when power voltages exceed 
design limits.  The motor and interface silicon chips are low mass and heat 
FAST.

Ted's "doubled drives" will increase reliability for mechanical drive failures 
(in my experience, mostly head crashes and plowed oxide on the disk platter, 
followed by spindle motor failure).  A good idea ... before 2020.

However, as a chip engineer, I prefer to leave the motors in the fab, moving 
wafers and attaching chips to packages long before they leave the factory and 
enter my door.
I have problems enough with cooling fan failures.

Still ... oxide on a platter can store "magnetic pixels"
as densely as a microscopic read head can find them, so the same lithography 
techniques that make small transistors can alos make very small read heads and 
very uniform oxide platters.

I presume disk drives will continue to pack more bits onto a square millimeter, 
enabling more bits per platter and drive.  Above 4 terabytes or so, disks win 
cost per bit (perhaps not cost-per-bit-YEAR or bits-per-WATT), so they have a 
place in many computer systems. 
But not for long.  Moore's Law continues to chew up legacy systems and digest 
them into microchips.

---

As for me, I use big magnetic platter disks for system backups, hundreds of 
soft-linked rsync images.  For now.

I will do this until somebody invents a "write once"
optical tape backup drive using inexpensive, indelible optical tape, 
solid-state-steerable UV lasers, and submicron "bit pixels".  A 5μm thick, 5mm 
wide, 200 meter length optical tape spool/cartridge with 1μm hex-grid pixels 
and 2x redundant coding might store
5 terabytes of backups in a plastic cassette the size of an audio tape 
cassette, with similar prices. 

A few pennies more for a megabyte flash "index" chip in the optical tape 
cassette, with RFID-style encrypted readout, so you can scan a drawer of 
optical tape cartridge indexes to find the data you want.

Platter drives will eventually follow DOS into history's dust bin. *

Keith L.

* (my hands first wrote "dust bunny", which is apropos of my sloppy information 
management processes).

-- 
Keith Lofstrom          [email protected]

Reply via email to