Re: reiserfs empirical study (very long)
Geert Uytterhoeven wrote:
>
> On Wed, 30 May 2001, Andrew Sharp wrote:
> > The big endian patches change the code to use little endian ordering
> > for all on-disk structures. IMO this is a mistake, and certainly
> > costs a dear performance penalty, because on big endian processors,
> > this method requires converting endianness both ways (reading and
> > writing) for all meta data. I submit that there is little reason
> > for this, and the performance cost is not worth the very dubious
> > feature of having the file system be moveable to little endian
> > systems, like x86. Note that except in few cases, the disk labels
>
> We had the same discussion many years ago about ext2fs, and a few years later
> about XFS. In fact m68k and ppc used to have a big-endian ext2fs.
>
> Now ext2fs is defined to store metadata in little-endian order, and XFS to
> store metadata in big-endian order. This was done for interoperability reasons.
>
> Since people do want to exchange disks between machines, the alternative was to
> support both endiannesses. In fact m68k did have a bi-endian ext2fs for a
I would actually like to hear more about these discussions. Are
there any archives? Are they too old? Geez, if some silly person
wants to take a disk from one machine to another, well, that is what
vfat is for, no? ~:^)
> while, which supported both little and big-endian ext2fs. However, this was
> even more costly because all byteswapping decisions had to be made at runtime.
> Compare this to the static case, where the compiler can optimize all
> byteswapped accesses.
>
> > alone would prevent this. I would very much like to see some endian
> > patches that don't have this affect. I believe that the large file
> > I/O performance and large directory tree copy performance would show
> > a definite increase. It may be too late now.
>
> Until you can prove there's a significant performance difference, I'm afraid
> Reiserfs will stay little-endian...
Actually I think it's already well understood, much as I would love
to quantize it. However, as processors get faster and faster, it
becomes less and less costly, and therefore less and less
important. It is significant on these aging and quite slow powerpc
chips. But probably extremely difficult to even measure on 600 MHz
G3s. Which will be obsolete in a couple of weeks. Hence, it will
be hard to get anyone to care enough to do the not-insignificant
work. You 68k guys would probably benefit greatly, however ~:^)
There are probably some subtle issues regarding this that I don't
know or remember. I shall endeavor to find out.
a
Reply to: