>>> Lars Marowsky-Bree <[email protected]> schrieb am 22.04.2013 um 16:46 in 
>>> Nachricht
<[email protected]>:
> On 2013-04-22T15:18:20, Ulrich Windl <[email protected]> 
> wrote:
> 
> > That's where OCFS2 comes in: When hosting Xen-VM images, multiple (we just 
> use the trivial two-node cluster) hosts may acess one LV (i.e.: one 
> filesystem), while access (if OCFS2 does allocate the data in big extents) to 
> specific areas is mostly local to one node.
> 
> Sure. Yes, this thin-provisioning is one of the most important features
> in OCFS2 for the virtualization use case.
> 
> > I don't know how mirroring works at the moment, but I guess one node
> > does the mirroring job for a whole LV. It would be probably better if
> > the mirroring of one LV could be distributed among nodes so that each
> > node would mirror the data it accesses. Then when idle, some task
> > could do the rest (that is out of sync, if any).
> 
> No, actually, all nodes write directly to disk. But access to the
> mirrorlog has to be synchronized, and the locks to invalidate read
> pages so that no stale data is presented.

Hi!

I'm not wanting to discuss this eternally, but from SLES11 SP1 to SLES11 SP2 
for MD-RAID the default bitmap chung size changed dramatically: Where some RAID 
used something like 200 pages for the bitmap, it now uses 3 pages (when you 
recreate the RAID, of course). Naturally the granularity of much coarser now, 
but maybe considering TLB this is more efficient on today's hardware. Maybe a 
similar change would be beneficial for LVM mirroring...

Regards,
Ulrich

> 
> > > At least not for released code, no, this is not the case.
> > > ftp://sources.redhat.com/pub/lvm2/ - last release is from October. And
> > > it doesn't improve this much at all.
> > Specifically I meant "mirrorlog=mirrored" described for RHEL 6.1, vs. 
> SLES':
> > 
> > # lvcreate --mirrorlog mirrored -m 1 --name cLV1 -L 8G cVG
> >   Log type, "mirrored", is unavailable to cluster mirrors
> > 
> > Of couse I couldn't check from the docs whether this would work for RHEL.
> 
> The mirrorlog=mirrored works only for local mirrors with RHEL 6.1 too,
> just like with SLES so far.  The cluster functionality is not present
> upstream:
> http://article.gmane.org/gmane.linux.lvm.devel/4620/match=mirrored 
> 
> Dong Mao from the SLE HA development team developed the necessary
> patches to cmirrord to allow this to work in a cluster context
> (available with SLE HA 11 SP3, feel free to test it as part of your beta
> program participation).
> 
> See http://comments.gmane.org/gmane.linux.lvm.devel/8891 how much
> response this generated from the upstream/RHT development community.
> 
> Seriously, you're misdirecting this particular rant.
> 
> 
> 
> Regards,
>     Lars



 

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to