Re: 2.4 kernel tweaking

[Harald Staub <staub@switch.ch>]

> > > I've also experimented on this for quite a while and found that only
> > > ext2 (or other non journalling fs) will allow the disk to spin down
> > > with the help of noflushd and setting noatime.
> > > I found no journalling fs which works. If you do please let me know
> > > how.
> >
> > XFS turned out to allow spin down of HD together with noflushd. However,
> > for kernel 2.4.18 it requires a kernel patch. The debian xfs kernel patch
> > packet worked perfectly.
> >
> > So I have ext2 on my root partition, everything else on XFS.
>
> Thanks, but XFS turned out to be quite slow on my laptop and as I only
> have  4GB HD fsck doesn't take too long so i'll stick with ext2.


This is probably stupid, but I am quite happy with the following 
"solution". I hacked the source. The 5 second commit interval is hardcoded 
in

fs/jbd/journal.c:

       journal->j_commit_interval = (HZ * 5);

I changed 5 to 300. What I think this means is that I may loose 5 minutes 
of work, which is acceptible for me, but crash recovery is fast. Since I 
have crashes only when experimenting with the kernel, I think this is not 
as bad as it might look at first glance. But, you know, no guarantees of 
mine, your filesystem might become garbled, your hard disk might become 
fried!

Harry