Re: ext3 sucks

[Joshua Baker-LePain <jlb17 duke edu>]

On Tue, 25 Oct 2005 at 10:12am, Kostas Georgiou wrote

> On Mon, Oct 24, 2005 at 04:19:30PM -0400, Joshua Baker-LePain wrote:
>
> > If there are any more suggestions/tweaks/patches, I'd love to hear 'em.
> > Thanks again.
>
> With 3ware cards i get slightly better results (for our workload) in RHEL4
> with the following settings:
> vm.dirty_expire_centisecs = 1000
> vm.dirty_writeback_centisecs = 500
> vm.dirty_ratio = 5
> vm.dirty_background_ratio = 10

These numbers had no effect for me, but I'm likely to play with 'em a bit 
more.

> Have you tried the kernel in the following bugzilla? I didn't
> had the time to test it but it seems promising.
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=156437
> (Writing large file to 1TB ext3 volume sometimes very slow)

I *did* try that kernel, to no improvement.

> You might also want to monitor this bugzilla entry if anyone finds
> out whats wrong with ext3 ad 3ware cards it will show up there:
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=121434
> (Extremely high iowait with 3Ware array and moderate disk activity)

Yeah, I've been watching that bug a while too.

Of note, I just tried ext2.  On a single hardware RAID5, it matched XFS in 
read speed and nearly matched it in writes.  On a software RAID0 across 2 
hardware RAID5s, ext2 was still not as good as XFS, but far better than 
ext3 (well, in writes):

     write   read
     -----   ----
ext2    81    180
ext3    34    222
XFS    109    213

Does that help narrow the problem down and/or suggest any more solutions? 
I just ordered more disks to put inside these beasts as on OS mirror drive 
(in case I go with the all software RAID50 solution), so maybe an external 
ext3 journal would help?

--
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University