> But that's not all that "stability" means. If you have hardware that
> makes Linux happy, linux is extremely stable. The standard fix for an
> NT machine is to reboot. Linux machines, on the other hand, can be up
> and running for hundreds of days at a time.
The problem is that we've had to reboot Linux on more than one occasion when
the X server that worked perfectly the day before and the day after chose to
crash the entire OS by screwing up the video. Granted, some Linux machines do
run unperturbed for long periods, but, then, we've seen that with every OS
we've ever used. In some circumstances, nothing ever happens; in others, uptime
is periodically punctuated with unpredictable downtime. Across the entire
population, Linux may average less crashing, but our concern is with what
happens in our shop, not elsewhere. The experiences of other shops does us
little good if we are unable to implement Linux here.
> Linux has benefits to make up for the holes in its hardware support.
> And the hardware support it has is really quite good, despite its
> problems. Things seem worse for Linux in part because it doesn't come
> pre-installed. If you bought a system from VA Research, I bet you'd
> have far less hardware troubles.
This is a valid point, certainly, but not entirely applicable in our case
because of the proposed application. Also, there's the question of whether the
hardware serves the software or the other way around. In a desktop application,
the hardware must be subordinate to the software. In an embedded industrial
application, this facile assumption is not as easy to sustain.
> One question, though: if you don't like Linux, and you don't see the
> benefit to your business, then why are you using it? Linux certainly
> isn't for everyone, at least not yet. If it doesn't satisfy your
> needs, maybe you should just stick to NT.
Good question. First of all, we are still experimenting with Linux. We don't
want to toss out a candidate which has not been evaluated fairly, and it takes
a lot of effort to do this. Second, NT is not suitable for the application
under consideration. We use NT where it's appropriate - desktop software,
solids modelling, C programming for DOS targets, etc. For process and machine
control, more determinism is required, even in soft real-time applications.
RTEMS is another candidate, and very likely a much better one, but it, too, has
the drawback of vague and questionable support.
I suppose that the real heart of the controversy from our point of view is
accountability. To the Linux community, it's okay if the user has to resort to
uncertain means of support such as the 'Net, users' groups, and e-mail for
whatever assistance is needed to get the OS running or back up after a crash.
In an industrial situation, downtime is simply not acceptable, and support for
the components that we incorporate into our systems, software or hardware, must
be immediate and absolute. Linux cannot provide this. With the DOS-based
systems that we now use, we can provide this kind of assurance to our
customers. As noted in another posting, we are trying to improve our control
product, and Linux is one possible means of doing it.
Maybe....
--
David Fisher
Chief Engineer
Fisher Research Corporation
Rochester, New York
[EMAIL PROTECTED]
716 328 4230
fax 328 1984
--
PLEASE read the Red Hat FAQ, Tips, Errata and the MAILING LIST ARCHIVES!
http://www.redhat.com/RedHat-FAQ /RedHat-Errata /RedHat-Tips /mailing-lists
To unsubscribe: mail [EMAIL PROTECTED] with
"unsubscribe" as the Subject.
