MESSAGE
| DATE | 2016-04-08 |
| FROM | Rick Moen
|
| SUBJECT | Re: [Hangout-NYLXS] Fanless Desktop Solutions
|
Quoting Ruben Safir (ruben.safir-at-my.liu.edu):
> Agreed with all this, which is why I posted and they DO actively support
> Linux system, although their stock choice is Mint.
Again, the question is (1) how adequate is 'supported' if one is
eschewing out-of-tree proprietary drivers, and (2) are all the chipsets
comprising the unit _better_ than barely functional on Linux,
irrespective of whether using mainline or out-of-tree drivers? (I.e.,
are they actually good, or do they merely kind of work?)
For example, as I mentioned, my CompuLab IntensePC has two GigE ports,
both correctly stated to be 'supported' in Linux Mint. One of them is
an Intel NIC, and is a generally sound choice. The other, as I
mentioned, is a Realtek RTL8111F.
The Realtek RTL8111F NIC has a mainline Linux driver, r8169. It also
has a proprietary, out-of-tree Realtek driver, r8168. As of 2011, both
were in different ways a somewhat miserable experience in Linux:
https://unixblogger.wordpress.com/2011/10/18/the-pain-of-an-realtek-rtl8111rtl8168-ethernet-card/
Has that situation improved substantially in five years? Possibly.
Probably I'll never know, because fortunately the server's intended
deployment requires only a single NIC, so I'll use Intel NIC#1 and
ignore Realtek NIC#2, _but_ past experience suggests that components
from Realtek generally should be avoided if possible, and my guess is
that the RTL8111F carries with it core suckiness that cannot ever be
fixed by a sufficiently good driver.
So, my attitude towards the word 'supported' in this context is that
it's heartwarming and will, along with $2.25, get you a ride on Muni
(https://www.sfmta.com/getting-around/transit/fares-passes), but is not
to be confused with real information about the quality of hardware and
Linux drivers.
Some of the CompuLab hardware choices have feet of clay, and not just
the Realtek RTL8111F one. You get similar badness with many of the
hardware OEMs, sometimes better with particular product segments and
sometimes a lot worse. With my old employer VA Linux Systems, you could
reliably assume that there were no crappy component selections and all
parts would have excellent, in-tree open source drivers -- and, in this
regard, CompuLab is just another hardware OEM that is not a Linux
company.
They can talk all they want and be 'open to discussions' all they want,
but the fact is that many of their component selections are questionable
for Linux use (despite the 'supported' language) and would have been
automatically avoided by any Linux hardware firm.
Don't get me wrong: I like them, and I happily bought an IntensePC over
all other competition -- but some of their hardware component judgements are
notably bad and have questionable Linux friendliness.
> But specifically I wanted to address the money issue. They might be a
> tab expensive, but they truly give so much more than a standard desktop.
I like to say (1) Does a unit achieve my minimum standards of quality?
(2) If so, how many years of useful economic life am I likely to get out
of it?
The answer to question #2 lets you divide the total acquisition cost by
number of years of useful economic life, yielding amortised cost per
year of useful economic service.
A unit that is likely to give useful service for ten years and cost
$2000 is cheaper than one you'll want to replace after five years but
costs $1200.
The trick is to judge in advance what's likely to give ten years of
useful economic service. Fortunately, I happen to be good at that.
It's called 'learning' and 'experience'.
_______________________________________________
hangout mailing list
hangout-at-nylxs.com
http://www.nylxs.com/
|
|
