On Thu, 7 Sep 2000, Greg Kettmann wrote:
> Question 2 - This is more a request for information. I've several high
> level customers who need statistics on Linux. We (us here and the Linux
> community in general) have done a good job of making Linux a viable
> alternative. We've got their attention. But now I need the next
> level. All that upper level managment junk like ROI (Return on
> Investment), TCO (Total Cost of Ownership), MTBF (Mean Time Between
> Failure), etc. I've read every article on the subject but am having
> trouble finding hard numbers, particularly on MTBF or reliability. My
> favorite source of good articles is OSOPINION. Does anyone have any
> suggestions on where I might get more information?
>
> Thanks again. GGK
>
>
Yea, they like it, but most of those statistics fit with Mark Twain's
"Lies, Damn Lies, & Statistics." They especially require a good
understanding of the environment (i.e. MTBF on a web farm serving
static pages can be high, but what's the MTBF of the controller of a
numerically controlled machine that sits on the floor of the
production line, in the dirt?). I don't have any articles, but the
below would be my reasoning:
TCO: this is acquisition cost+support costs (training is included in
support costs). At minimum, you can say it's better than Windows
because your acquisition cost is $0. Actually, it can be better,
because you choose your support costs (you're not tied to a single
vendor). Some will argue that the cost of training (retraining) and
cost of admins needs to be included. I say, however, anyone who
thinks you can run a Windows network (or any network) without trained
admins & without training your employees (including the training of
the month on whatever is the latest MS feature) is deluding
themselves. Also, the same argument applies against the proprietary
Unices, plus the fact that Intel HW is cheaper (although you may pay
in reliability, see below).
MTBF: Apples to Apples is difficult, since the proprietary Unices
have a tendency (actually, they always are) to be built on much higher
end HW, and it depends on the application environment. Assuming intel
HW, running the same tasks, comparing to Windows:
1. You only install what you need (anything not installed doesn't
break).
2. You can do most fixes without a reboot (I consider planned
downtime part of downtime. Regardless of why it's down, it's not
available).
3. Clean separation of userspace from kernelspace. Thus, apps will
kill themselves, not the system.
4. OS fixes happen FAST (quickest I've seen is Alan Cox posting the
problem & fix in the same message, more typically, look at the
bugtraq postings, and vendor patches - also Security Focus did some
statistics on this, which had Red Hat faster than MS or Sun (they
didn't look at any other distributors, or we probably would have had
even better times).
5. Clustering that works (OK, it's us, although I don't want to
exclude anyone else, I just am less familiar with the other
clustering). This really depends on their app, but this easily adds a
9 to the "9's" of reliability.
Having said the above, measuring some MTBF would be something great
for some large Linux Company (hint, hint) to do some testing in some
kind of open lab that they support.
jeff
------------------------------------------------------------------------
Jeffry Smith Technical Sales Consultant Mission Critical Linux
[EMAIL PROTECTED] phone:603.930.9739 fax:978.446.9470
------------------------------------------------------------------------
Thought for today: file signature n.
A magic number, sense 3.
**********************************************************
To unsubscribe from this list, send mail to
[EMAIL PROTECTED] with the following text in the
*body* (*not* the subject line) of the letter:
unsubscribe gnhlug
**********************************************************
