Re: how to make Debian less fragile (long and philosophical)
I have not run 10,000 machines over my career, and I have seen more
than one failure. I have run probably not even 50 machines, and
five or six times I've needed statics. In my experience, then, the
ratio is closer to 1 in 10.
If you want to try and analyze this a different way, estimate what in
your opinion is the mean time to failure for a Debian machine. How
many days do you think would pass, with normal upgrades and normal
operation, before some kind of error, hack, mistake, or bug would
bring your machine down such that you needed statics.
What do you think? It's obviously more than 100 days (which would
have you doing it three times a year) and obviously less than 10000
days (which would have you doing it once every 27 years).
Maybe you expect to run into a problem like this, on a single machine,
once every six or seven years. That sounds pretty reliable, and puts
the mean time to failure at something like 2500 days.
Now learn why that is an unacceptable risk:
Let's say you are running 10 machines with a MTTF of 2500 days. You
will see a failure like this once every 250 days, or about once a year.
Now before you criticize me let me state the obvious: all of these
numbers are totally fabricated, except the 1:10 report which represents
the single date point of my own experience. The mathematical MTTF
model I am using is completely bogus. It's not going to work out
exactly like that, and I can't make any claims as to what the real
failure is, and these calculations are absolutely not accurate.
What I intended to show you was that even if Debian is pretty reliable,
it is hard to be so reliable that good, durable recovery tools are
not still important. When you sample over a large enough number of
users, or take the case of an organization with a large enough number
of machines, the low failure rate compounds and suddenly someone
loses something important.
The identical arguments are used for backups. Some people don't make
backups because they think their machines will only fail once ever
seven or eight years. Not all systems that require backups require
live recovery tools, but in many cases it's important. So live
recovery is not as important as good backups, but it almost is.
I think that statics should be the default, that every user should get
for their own good. Intelligent users, who are damn sure what they
are doing, and know that their system is not important, should have
the option of not installing it--but that shouldn't be the default.
Justin
On Wed, Aug 18, 1999 at 10:13:50AM -0700, Steve Lamb wrote:
> Wednesday, August 18, 1999, 2:24:59 AM, Marek wrote:
> > Oh, not again. That's pure demagogy (sp?) - it DOESN'T matter that they will
> > be useful in 10 out of 1000 cases, can't you understand it?
>
> 10 in 1000? I see less than 1 in 10,000. Meanwhile in the other 9,999
> cases it is wasteful.
>
>
> > It's the same as if you said that mounting an air-bag in a car in a *VAST*
> > >>>MAJORITY<<< of cases they are wasteful. You have a choice - DON'T buy an
> > air-bag, DON'T install those binaries. Point.
>
> Actually, air-bags were legally required to be installed for a number of
> years. People didn't have a choice. In some locations they still do not.
> Nevermind that they are detrimental in some cases. Now there are laws
> requiring people to specifically ask for airbags not to be installed on the
> passenger side because they have caused fatalities in children in car seats on
> the passenger side.
>
> Lovely, isn't it.
>
> You have a choice, if you want statically linked binaries, *COMPILE THEM*.
> It is my opinion that if a person needs such protection they had also better
> have the expertise to compile what they need and be willing to *USE* that
> expertise. If not, go to NT, we don't need you here.
>
> > Yeah, yeah. There is memory cost, of course, and a huge one - but they will
> > be used in such rare cases that it in *VAST* >>>MAJORITY<<< of cases doesn;t
> > matter.
>
> Uh-huh, each time you run those common commands it incurs extra memory
> usage. Gee. My laptop will love you. Geez.
>
> --
> Steve C. Lamb | I'm your priest, I'm your shrink, I'm your
> ICQ: 5107343 | main connection to the switchboard of souls.
> -------------------------------+---------------------------------------------
>
>
>
> --
> To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
>
Reply to: