Re: TAKE - fix raid5 journal problems

[Russell Cattelan <cattelan@xxxxxxxxxxx>]

Aggh, sorry I didn't notice this sooner.... haven't had much of a change 
to look
at all the takes lately.

I messed around quite a bit at one point trying to convince MD to take 
full page
requests, but always eventually end up a with a corrupted log. If I 
remember right
one of the log sectors would end up on a sector +1 to where is suppose 
to be.
I finally gave up since I wasn't able to measure much performance 
difference, but
I did have a limited # of fc disks so no idea if that will hold true for 
larger stripes.


Steve Lord wrote:

> Backing out an optimization, this code appears to make raid5 misbehave and
> fill the log with garbage. This caused recovery to fail, and sometimes a
> clean mount to fail. I am unable to replicate these problems once the code
> is removed. There will be a performance impact from this, but I am not 
> sure how much. Hopefully in the long term we can find what is going wrong
> at the md layer and put this optimization back again.
>
> Steve
>
> p.s. for those of you not following the discussion recently, the best
> performance for raid5 is an external log on a non-raid5 device, raid1
> if you want to maintain the backup copy of the data.
>
>
>
> Date:  Mon Jul 23 15:54:45 PDT 2001
> Workarea:  jen.americas.sgi.com:/src/lord/xfs-linux.2.4
>
> The following file(s) were checked into:
>  bonnie.engr.sgi.com:/isms/slinx/2.4.x-xfs
>
>
> Modid:  2.4.x-xfs:slinx:99431a
> linux/fs/pagebuf/page_buf.c - 1.95
> http://gibble.americas.sgi.com/cgi-bin/cvsweb.cgi/slinx_2.4.x-xfs-nodel/linux/fs/pagebuf/page_buf.c.diff?r1=text&tr1=1.95&r2=text&tr2=1.94&f=h
> http://oss.sgi.com/cgi-bin/cvsweb.cgi/linux-2.4-xfs/linux/fs/pagebuf/page_buf.c.diff?r1=text&tr1=1.95&r2=text&tr2=1.94&f=h
>         - Backing out the change which made pagebuf do larger I/O on md devices
>           in some cases. This will increase the cpu overhead in this case, but
>           fix the log on these devices which was broken.