On Tue, 30 Apr 1996, Michael H. Warfield wrote:
[...]
> > >We got a hard lesson in this recently due to a freak power failure. An
> > >entire rack of Unix machines lost power. Three RS/6000s of various
> > >models all came back up perfectly as soon as power was restored. Two
> > >SPARC 20 machines didn't fare quite so well... their internal drives
> > >were fine, but their external drives were horked. And the PC running
> > >Linux? It took 15 minutes to get its filesystems back into a runnable
> > >state.
>
> > Most people that I know would not run Linux in production. SCO or Unixware
> > make much better choices.
>
> I run Linux and SCO systems both in a personal environment and
> in production servers. For reliability, Linux beats the living bejesus
> out of SCO hands down! I've run SCO systems for years. I've got multiple
> flavors of SCO Xenix, SCO UNIX and SCO ODT up and down the spectrum.
> I've done device drivers and system applications for SCO Xenix, SCO UNIX,
> SCO ODT, and Unisys UNIX. My Linux systems are currently under more load
> than I would ever trust my SCO systems with and yet my SCO systems "go down"
> more often.
>
I can confirm : I got Linux then SCO on exactly the same machine (Intel).
Linux remains up for 4 months with great perfs. SCO ODT3 is never up more
than 20 days long before a crash with poorer perfs. (Note that is was the
same machine *physical* machine. not two ones).
> > >The reason for this isn't software, it's hardware. The power supplies
> > >are designed to protect not only the hardware, but also the filesystems.
> > >When a power failure happens, the power supply notifies the operating
> > >system kernel, which immediately initiates a clean shutdown. The power
> > >supply has enough onboard capacitance to run the machine for the several
> > >seconds needed to at least sync the drives.
>
The RS/6000 machine electrical capacity allow more than 500ms of run after
power down. That's true. The new ones are more sensible.
Anyway, i remember a spent night due to the journaled FS on AIX3.1 : the
corrupted log corrupted in turn the filesystem !!. (It is a corrected bug).
> Gee... I've got that on all of my X86 systems with Linux. What
> did you do wrong??? And if you think those SPARCS have that much capacity
> in those power supplies - you are dreaming!
>
> > Without a UPS your asking for it.
>
The choice of UPS *IS* important. I've got a machine with more power fault
when plugged on a UPS than in the line.
> No bout a doubt it!
>
> > >PC hardware doesn't have this sort of support. Remember, it was
> > >designed with the DOS FAT filesystem in mind, which isn't sensitive to
> > >system states the way UNIX filesystems are. So whenever power is lost,
> > >the system loses state and the filesystems get horked.
>
THAT'S FALSE !!! I worked, long time ago, on DOS and files are written only
when you close the file (wich is a looooooonnnnnnnnnnggggggggggg operation).
During the time the file is open, the file is zero lenght, blocks are
allocated (and lost on power fault. (or system crash)).
I loose mountains of data with DOS, even on files not concerned in the
operations. I remember guys saying : 'Unix is a bad system : you can't
power it down without entering a special command or you will loose ALL
your data !!'.
(Never forget DOS comes from QDOS : Quick and Dirty Operating System).
> > __________________________________________________________________
> > Copyright 1996 (c) Penn Jennings
-------------------------------------------------------------------------------
S Y N C H R O N I X S.A.
Avn des ANDES, Bat. LE CEDRE - 91952 LES ULIS - FRANCE
Tel : +33 1 64462626 - FAX : +33 1 64466976 - Internet : Synx.com
Remy NONNENMACHER - APAV Dpt. (remy @
synx .
com)
#include <Following-tags-are-my-own-and-not-of-my-employer>
#include <Copyright-GNU-like-but-excludes-M$-network(s)-and-affiliates>
----
"My uncle and I entered a better life" - Alfonse Allais
(When his rich uncle died).
References:
|
|
