Ext3 users - Oct 2003 - ext3 on RAID1 gains very little?

I have a software RAID1 underlying the ext3 filesystem on my laptop.

This was using xfs, which took advantage of the mirroring very well, but
that didn't leave me feeling very comfortable about the reliability of
the system; no fsck was not a nice feeling.

Now that I have moved to ext3, I very seldom see the mirroring being
taken advantage of, and the system responds a lot more slowly if there
is disk load present.

This is especially visible when starting a session, which loads three
major applications from disk at once. Under XFS, I got two to five meg a
second of each drive during that.

Now, I see the first drive pegged at around 1 meg a second, the second
drive almost completely idle, and it takes a fair bit longer to load the
software.


Is this expected, or is there something odd about my setup that is
hurting performance?

If it is expected, is this a design limitation of the filesystem, or a
simple matter of programming to avoid excessive locking (or whatever)?

       Daniel

-- 
If you lie to the compiler, it will get its revenge.
        -- Henry Spencer

under what journaling mode you're running ext3 - data=journal or
data=ordered ?

note that if you have mirroring under full journaling, every write is
going to take 4 write i/os - 2 mirrored writes for the journal and 2
mirrored writes for the actual data.

but you seem to have this high load when starting a session and may be the
reads are not giving the full bandwidth. in raid-1, if the reads are
sequential they are directed to only one mirror. that may be the reason
why the second drive is almost idle.

Vijayan

On Fri, 24 Oct 2003, Daniel Pittman wrote:
> I have a software RAID1 underlying the ext3 filesystem on my laptop.
>
> This was using xfs, which took advantage of the mirroring very well, but
> that didn't leave me feeling very comfortable about the reliability of
> the system; no fsck was not a nice feeling.
>
> Now that I have moved to ext3, I very seldom see the mirroring being
> taken advantage of, and the system responds a lot more slowly if there
> is disk load present.
>
> This is especially visible when starting a session, which loads three
> major applications from disk at once. Under XFS, I got two to five meg a
> second of each drive during that.
>
> Now, I see the first drive pegged at around 1 meg a second, the second
> drive almost completely idle, and it takes a fair bit longer to load the
> software.
>
>
> Is this expected, or is there something odd about my setup that is
> hurting performance?
>
> If it is expected, is this a design limitation of the filesystem, or a
> simple matter of programming to avoid excessive locking (or whatever)?
>
>        Daniel
>
> --
> If you lie to the compiler, it will get its revenge.
>         -- Henry Spencer
>
>
> _______________________________________________
> Ext3-users mailing list
> Ext3-users@redhat.com
> https://www.redhat.com/mailman/listinfo/ext3-users
>

On Thu, 2003-10-23 at 21:09, Daniel Pittman wrote:
> I have a software RAID1 underlying the ext3 filesystem on my laptop.
> 
> This was using xfs, which took advantage of the mirroring very well, but
> that didn't leave me feeling very comfortable about the reliability of
> the system; no fsck was not a nice feeling.

I think you may be misunderstanding something about how XFS works. 
"fsck" doesn't do anything on XFS volumes because its journaling
layer handles
filesystem integrity checks at startup.

However, if you still want/need to manually repair an XFS filesystem you can use
the
"xfsrepair" tool, which has been around a long time and works very
well.