&don_hat($devils_advocate);
[ Monday, January 24, 2000 ] Thomas Waldmann wrote:
> > I believe that Ingo realizes that no amount of last-minute cram-testing
> > is going to be a real factor, it'll come down to Linus's feeling of
> > how well reiser, ext3, and 0.90 raid play together.
>
> Well, although all of reiser/ext3-fs and raid are very fine stuff, I would
> definitely vote for SW-RAID to be preferred (if they can´t be included
> alltogether due to incompatibilities)
Thankfully, they can and will be living happily together, so the
rest of this email is now deemed Moot(tm).
Resier and ext3 have the same issues (buffer-cache-wise) with s/w raid
(raid5 and resync in particular) so I'm not sure that it'd make sense
to kick out 2 journaling fs's (that can and most likely will be used on
99.999% of desktop-and-above Linux machines) in favor of s/w raid that's
likely to stay in the minority userbase of Linux (regardless of how much
we like it).
Also keep in mind that if the issues were to stick around, this will
most likely keep XFS out of the kernel as well... SGI doesn't need any
more obstacles to that port than they've already got :)
> - because it´s been there for a much longer time
True.
> and because it gives important functionality and saves $xxx(x)
> for not having to buy a HW raid controller. Ext3 and ReiserFS "only" give
> faster recovery times and better performance and are not as widespread in
> use as RAID 0.90 is (IMHO).
For 24x7-ish machines with TB-range filesystems, you're talking *lots*
of time in fsck... for machines where minutes of downtime mean piles
of cash lost in transactions (amazon, cisco's site, etc)... making
a machine that can reboot in 30 seconds doesn't mean much if you're
gonna be doing hour-long fsck's :)
"not as widespread use" may or may not be true... ext3 is in the rawhide
kernels and gets used on every laptop in my LUG... compared to only 3 or
4 of us that are using the s/w raid capabilities. Journalling (esp. when
perf isn't hurt, which somehow sct's done an amazing job with) is just
more applicable to a much wider range of machines. s/w raid has its place
to be sure, but from a Big Picture view journaling has to be considered
a priority.
> I made quite some 0.90 raid systems (about 10 ?) for customers, but NO 0.4x
> system at all, also NO reiserfs and NO ext3fs system. So my impression is
> quite opposite to yours.
Do you really believe after their use has been diseeminated down to the
major distributions that s/w raid will get more use than journalling?
With sct's performance numbers being what they are (and resierfs being
b-tree based), sticking with ext2 is a loss for new installations.
That you don't have enough confidence in the code bases at this time
is understandable, but long-term it's hardly realistic to say s/w raid
will dominate.
For a good break-down of what we should expect, we can look at WinNT
(like it or not :) which has s/w raid *and* a journalling fs (NTFS).
I'd dare say it's painfully obvious that NTFS gets much more use than
their s/w raid capabilities.
Yes, this whole thread is moot since the conflicts {are,will be shortly}
resolved. :) But it seems obvious to me that a journalling fs on top of
large s/w raid (or any large fs) just makes the most sense. With this
hope soon becoming a reality, life in 2.4 land will be a happy one,
regardless of whether I have to do a patch -p1 to get my kernel src base
the way I like it :)
James
--
Miscellaneous Engineer --- IBM Netfinity Performance Development
