[tech] A question of File System

[Nick Bannon]

On Mon, Apr 22, 2002 at 04:39:07PM +0800, Grahame Bowland wrote:
> I've never been sure if it's journalled or not. NT4 seems to need a fsck
> but 2k on NTFS doesn't seem to.

I believe what it has are background fscks, like VMS.

It's certainly a practical way to get your machine up and running as
fast as possible, but it introduces Hard Problems (tm) about what to do
if you do happen to find corrupt data after your applications are up
and running and serving it to the outside world, storing it in your
databases, etc.  Reboot, perhaps? ::-)

There's quite a lot of l33t filesystems these days - perhaps ext3 +
Daniel Phillip's htree directory indexing patch. Constant time
directory operations in measured in the tens of microseconds, million
file directories, etc. ext2 is a nice base to build things from, but
updates to the core kernel code will be conservative, slow, and
massively tested by Ted T'so. ext3 will work in multiple modes - full
data journalling, the metadata-only journalling that most journalling
FS'es use, an ordered mode that guarantees no stale file blocks after a
crash and journalling to a separate disk (eg massively fast NVRAM).

Another of Daniel Phillip's plans is the Tux2 filesystem that'll make
journalling obsolete and snapshotting easy. Don't think there's been
much progress recently.

ReiserFS is fine, though I don't really notice any performance
difference. I use it on my laptop and it was certainly handy that it's
resizable and easily movable - I was able to shuffle it down the disk
by 50MB to avoid some bad blocks. Much is planned for it, some of it's
even sane.

XFS is nice, at least on SGIs. "I'd like 30Mbps guaranteed bandwidth to
video spool, please".

JFS is also mature and featureful.

Nick.

-- 
   Nick Bannon   | "I made this letter longer than usual because
nick-sig at rcpt.to | I lack the time to make it shorter." - Pascal