Sam Hill wrote:
On Mon, 04 Mar 2019 19:11:45 -0800, T wrote:

On 3/4/19 6:38 PM, Sam Hill wrote:
On Mon, 04 Mar 2019 17:55:57 -0800, T wrote:
For those that need to run 24/7, doing a reboot in the night clears
them up.
Think about what you just wrote: "For those that need to run *24/7*,".
How does a reboot fit into that?
Who is processing something 100% of the time?

The remote server that sits waiting for world-wide salespersons to log
business transactions... Geez, think beyond the mom-n-pop people you serve
for once.

You pick a stop when
nothing is expected to be happening, like when all the remote loggers
have finally gone to sleep, etc.. And just let everyone know when the
reboot will occure.

What if the loggers are in Denver, Dubai, and Dingzhou?

My computer has been running since the last neighborhood power failure,
67 days, 12 hours, 10 minutes ago.
A Windows machine?

No. (Now you're going to complain that 'this is a Windows group' - just
know I was picking on your silly comment about what "24/7" is.)

That is the longest I have seen! 12 days and one customer could not
type, or send eMail, and Chrome's windows shimmered.

"A Windows machine?" g

Maybe it's okay in your little world that not being able to run more than
a few days is sufficient.

By change do your have hibernate/sleep set?

You'd be running a redundant system with some sort of
fail-over, such that with a pair of machines, one could
be rebooted while the other takes over.

The only problem with schemes like that, is when they
get into a loop. One evening at work, a "pair" like that,
swapped roles every 90 seconds, affecting service, so I
had to page the on-call IT guy to put a stop to it.

Char could probably tell you how to set something like that up.

And there have been implementations like that, which are
run by idiots. My old ISP insisted on some sort of brain-dead
DNS scheme, where about 30 pairs of machines provided DNS.
And they would *reboot both machines in a pair* on the
same minute at night, knocking out DNS service to the customers
on that particular pair, for around 15 minutes. Talk about
strange designs... Why have redundancy, then reboot both machines
in the pair at the same time ? Boggles the mind.

Why not have... ten machines in parallel, and boot all
ten at the same time :-/ You know ya wanna.

My current ISP has "a single pair", for which the provisioning
is all hidden. (You cannot discover there are 30 pairs of machines
over time, by studying the DNS addresses it hands out over DHCP.)
You cannot see what goes on, inside the ISP building. I suspect
a kind of "virtualized" DNS scheme, which presents two addresses,
but has some unknown level of "proper" redundancy that we cannot
see. That setup has *never* gone down, that I could detect.

The Space Shuttle had five computers, and majority voter logic.
It ran two sets of software. It had a pair of machines on
each software set. And the fifth machine was some sort of
tie breaker. So there are people who have made reliable
solutions before, that actually work... They obviously
didn't hire any staff from my old ISP. If my old ISP
was running the Shuttle, they'd pick the landing cycle
as the time to reboot all five of them.

Paul