On Tue, Feb 26, 2002 at 12:00:35AM -0500, Pat Finnegan wrote:
gcc keeps on getting better, so that might reasonably
happen.
OK. I am no compiler expert, but I doubt that. GCC is a big monster.
Its main advantage is that it supports lots of target CPUs,
but that is also a problem. I think it will never be that good
like a compiler that targets only one CPU and is highly optimized
for that CPU. See the MIPS Pro cc...
Have you tried using Linux's LVM?
We are
using it on a production machine (dual PIII, 3/4GB RAM) with
450GB Hardware RAID and lots of our workstations.
Of course, it's not quite a filesystem, but if
combined with ext3,
it would probably be a pretty viable solution.
We are using the SGI XFS on the
LVMs. I didn't want to use ext3,
as it was marked highly experimental (At least more experimental
than XFS.) and it suffers from the same problems like ext2:
2GB file size limit, fix number of inodes, slower, ...
XFS has nice features like on line resizeing (A file system
_must_ be mounted to grow) and is faster then ext2/3, but the
buffer cache that comes with XFS is a bit, uhm, "strange"...
reiserfs needs to be tweaked to store files larger than 2GB
and has no quota support. (We need quotas and I consider a
*ix file system that has no quotas as incomplete.)
Our test of IBM JFS was a debacle.
Anyhow, why do you need to have just one
filesystem unless everything is just stuck in one directory?
Did I say that I put
everything in one FS?
One advantage of AdvFS is the distinction from file system storge and
file system namespace. You put a file domain on a disk partition (or
LVM). Then you create one or more file sets in this file domain. The
file set is the actual file system, that gets mounted. The storage
space that this file set needs, both file data and file system meta
data, are allocated from the file domain. If you have multiple
file sets in a file domain, they share the available storage space.
Of course you can put quotas on file sets to limmit the total amount
a file set can allocate below the free space of the file domain.
e.g. You create a file domain for /var and have distinct file sets
for /var, /var/spool, /var/tmp, /var/adm... You don't have to waste
the max. amount of data that a singe file system may need for every
partition. You create a file domain, that is big enough to hold
the average space for each file set. If a file set needs temporarily
more space, it gets this from the "overhead" space of the other
file sets.
--
tsch??,
Jochen
Homepage:
http://www.unixag-kl.fh-kl.de/~jkunz/