Re: reiserfs empirical study (very long)
Hello!
Ethan Benson schrieb:
> yeah well life sucks, deal with it.
Done :-)
> > > If the system disk of my PPC box dies, I'll be happy to restore it from backup
> > > on some other machine...
> >
> > That´s no problem with the idea above. For restore speed is no need, so
> > it does not matter if i loose 1% filesystem performance for backups or
> > repairs. But i think if a linux-box runs low on memory, 1% faster
> > disk-I/O can make a lot of difference.
>
> maybe, it all depends how much work it is to either maintain a big and
> little endian filesystem driver or to make some mount time option to
> twiddle it.
I´m from the mac-side of live, i do not like options, i like smart
software that determinds the fs by its own.
I like options to force a program to do what I want, maybe doing stupid things.
So i prefer to let the kernel look at the partition-table (or another
place on the disk) and let the kernel decide to use little or big-endian fs-drivers.
So the kernel includes two fs-drivers for performance reasons, one for
little- one for big-endian fs and which one to use is decided at boot-time.
With this is it possible to have different endians on different partions
on the same disk.
(Does not makes a lot of sense to mix endianess, but it would be possible)
For new filesystems the prefered endianess is determined by the
architecture that runs mkfs. (Can be overwritten by options)
> that of course still won't help you (performance wise) if you move a
> disk from one endianess to another permanently. unless you run some
> most likely rarely used, bitrotted, buggy converter. (yes the ext2fs
> one was pretty sucessful, but i did have someone tell me it ate a
> filesystem or two...)
Well, I do not want to change endianess on an existing filesystem
(sounds very very dangerous to me :-))
But it would be very nice to put a disk from a crashed system into any
linux-box and repair the filesystem and let it run there until the
original system is replaced by a new one.
> its all tradeoffs and what the best ones are to make. in this case i
> suspect code simplicity and maintainability is going to win out.
That´s what I mean, a lot of work at the big-endian-side to support
software that was written with little-endian machines in mind.
But I think it will never happen, it is just a thought.
Bye,
Christoph
--
Dipl. Ing. Christoph Ewering C & E Informationsdienste GbR
0 52 54 80 68 66 oder 0173 566 266 1
eweri@cunde.de
Reply to: