Re: Why choose Debian on server
On Friday 04 January 2019 09:57:07 Curt wrote:
> On 2019-01-04, Gene Heskett <gheskett@shentel.net> wrote:
> >> Dear Ivan, no one said something about systemd, because the topic
> >> was discussed and closed on the list.
> >> Not sure about Gene, but I use good old sysv init and for Gods
> >> sake, no one thinks of running systemd on a firewall (I hope)
> >
> > Just one of the reasons I have 5 boxes here running wheezy yet, one
> > running jessie. But its an armhf, an r-pi 3b TBE, and it is also
> > behind dd-wrt. Perhaps I should watch the dd-wrt logs to see if
> > Ivan has come calling but no one answered the doorbell?
>
> You're running an obsolete release with no security support on five
> machines to obviate the eventuality of getting cracked by the Ruskies?
I'd be having nightmares if they were all hooked to the net with nothing
but a switch or hub, but they are not. dd-wrt is the guard dog, and he's
insatiable, eats that stuff up and never gets fat or leaves any
excrement/residue. Best kept security secret I know of. A little bit
like JoAnne Dow's pet dragon Mikey, who'se trained to answer the
doorbell. Cucamonga(sp?), where JoAnne lives, ran out of door to door
salesmen years ago.
Even if I was to update my stuff, I'd have to save out and reinstall a
3.4-9-rtai-686-pae kernel (which isn't in fact pae) because its an rtai
patched kernel which can run the base thread at a 25 microsecond
repetition rate with latency's in the 2 microsecond timing error at the
50% point in a histogram. Normal kernels are just a big splat spread
over 100's of milliseconds. Even worse if an nvidia driver is installed.
You cannot drive a stepper motor when you haven't a real heartbeat.
So later kernels have to be patched to even maintain a decent 1
millisecond servo thread rate, using an fpga card to handle the
microsecond critical stuff. And that raises the cost of an LCNC install,
depending on what you want the machine to do, by as much as $300 a
machine.
Because of poor kernel performance, there is no new full iso release of
LinuxCNC newer than an older wheezy. The LCNC buildbot has stuff for
stretch, but only for simulation, the realtime stuff just doesn't cut
it. For either jessie or stretch. My jessie install on an r-pi-3b is
bleeding edge and occasionally needs bandaids. Mostly avoided by my
pinning the kernel and its matching library, both of which have a
totally different numbering scheme from the normal x86 versions.
uname -a reports it as 4.4.4-rt9-v7+ #7 SMP PREEMPT RT. And a
latency-test report says I should move more of the not so time critical
stuff to an even slower thread, its taking the 1 millisecond servo loop
around 1.25 milliseconds to run. I have 90% of the hand controls running
in a slower 5 millisecond (200 hertz) loop. There are no hand cranks on
that 70 yo machine, motors driving ball screws have replaced all that,
so it has $20 encoder dials from mpja.com to move it by hand. Movement
per click adjustable starting at .0001" per click. All built by me.
Stephen R. has found the biggest timing problem with the later 64 bit
kernels and is directing the effort to get rid of that time wasting call
in favor of something faster, so I'm copying the linux-rt list to track
that effort. But this is already TL;DR. And I have to go see if my
missus is ready for some lunch.
Take care now Curt.
Cheers, Gene Heskett
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
Genes Web page <http://geneslinuxbox.net:6309/gene>
Reply to: