[conspire] Fwd: Another motherboard was _not_ burned out today

[Rick Moen]

Quoting Ross Bernheim (rossbernheim at speakeasy.net):

First, thank you for trying to help.

> We tend to think of AC power as being reliable and consistent. In many
> areas it is not.

Well, I know that, but West Menlo Park had always seemed a _very_ unlikely
example, because it's California and because it's the rustic suburbs.
We don't have heavy industry anywhere near (e.g., no switching on and
off of big motors), and almost never have lightning storms.  There are
places that are famous for power fluctuations such as Long Island
(mostly industrial interference) and northern Ohio (lightning), but
we're very unlike those.  So, I've been slow to believe that it's likely
a problem _here_.

I was concerned enough after a couple of the incidents I mentioned (the
ShuttlePC, the VA Research model 500) to buy the two power conditioner
units I mentioned.  I'd been slow to deploy them for reasons that might
bear some discussion.

One, I'd devoted some thought to making sure I was setting out to solve
the right problem.  Solving the wrong problem is almost ubiquitous in
IT, and a little caution goes a long way.  There were variables that
needed to be taken into account before jumping to conclusions about
cause.

Two, I had to carefully weigh well-intended but actually bad advice from
friends to deploy a UPS and/or MOV (metal oxide varistor) device.  An
MOV is so extremely slow-blowing that it's almost completely useless, as 
equipment will be destroyed long before it blows.  MOVs are common not
because they're good but because they're cheap.  A UPS is a more-complex
problem:  As a class, such devices are primily big honkin' lead-acid
battery packs (that must be replaced every few years as a recurring cost
equal to about 80% of your original purchase).

UPSes follow one of two basic design patterns.  The older one has the
battery _not_ inline to the downstream devices.  The newer one does have
the battery inline.  Neither typically provides meaningful power
regulation but the older type provides none at all.

The battery pack in a UPS, new or old style, can often be a single point
of failure.  It is also not unknown for a failing UPS to literally catch
on fire.

The proper use-case for a UPS is 'I am willing to complicate my system
and shell out money every few years in order to have continuous uptime
across power outages of up to half an hour or so.'  Personally, I really
could not care less about brief PG&E power interruptions on average once
every couple of years, and, ever since the California Electricity
Crisis during the summer of 2001 when I converted my server to journaled
filesystems, it comes back up automatically after a power outage.  So,
what UPSes are good for is exactly what I don't need.

So, no.  Absolutely no UPS, thank you.  Very wrong problem, very wrong
solution.

If I hear another friend recommend this particular bad solution, I may
need to FAQ the answer, because this conversation is getting old.


Three, I am habitually very careful to change only one thing at a time.
Otherwise, you get diagnostic madness.

Four, I had been just so extremely busy with a grueling job for the past
six years that I simply had no time or energy to deal with
infrastructure problems at home, so a number of lingering problems went
unaddressed for a long time.  Voltage regulation was one of them; my
degraded RAID1 mirror pair was another.


> And even at the best of times it is subject to sags, spikes, noise and other 
> problems.

Do you have relevant data?  I didn't -- other than the emerging pattern
at my house.

I've seen what happens when people go deploying technology based on
guesswork and random hypothesising.  It's bad.  I try to never be that guy.

> You taught me the importance of a good power supply many years ago. But I 
> also put at least a good UPS with MOV’s to handle spikes and filtering between 
> my computers and the AC grid. It has prevented a lot of problems.

I'm sure the quality PSU helped you quite a bit.  I have extreme doubts
that the UPS or (especially) the metal oxide varistor did.

Since you mention PSUs, I _would_ have swapped out the infamous VA Linux
Systems PSU for something better years ago, if that had been possible.  
Unfortunately, it is a specialised unit, and you cannot replace it with 
a regular marketplace ATX power supply of any sort.  What I _did_ do was
make sure I collected a large pile of spare VA Linux Systems bespoke PSUs, 
in case some of them died.  Hilarously, I'll end up having about six of
those PSUs, whereas I've reached the end of my supply of Intel L440GX+
'Lancewood' motherboards.

OTOH, it's 2015, and past time for all of that gear to become landfill.