WRITE_AND_VERIFY doubt

Flavio Junior billpp at gmail.com
Fri May 8 15:48:18 PDT 2009


Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r0905087_f.htm">HTML-formatted message</a>

On Fri, May 8, 2009 at 6:05 PM, Qi, Yanling <Yanling.Qi at lsi.com> wrote:
>  Flavio,
>
>
>
> I think I understand your configuration. You have two DS4700 arrays. You
> replicate your volumes between the two arrays. Both the primary volumes and
> the secondary (the replicated) volumes are visible to your Linux host. The
> replicated volume has write-protection. If the Linux host can not access
the
> primary volumes or a primary volume becomes secondary (due to role switch),
> you would like to start your read/write access to your new primary volume.
>
Yes Yanling, this is exactly what I have. Under dm-multipath i've eugenio
LSI RDAC (MPP) driver doing its work for multipath to same LUN's/devices.
Something like:
primary storage LU (path1, path2, path3, path4) -> eugenio RDAC driver
path1+path2+path3+path4 = /dev/sdb
secundary storage LU (path1, path2, path3, path4) -> eugenio RDAC driver
path1+path2+path3+path4 = /dev/sdf
/dev/sdb + /dev/sdf -> dm-multipath -> /dev/mpath/alias0
and "alias0" is my device to work to.
>
>
> First, the device-mapper multupathing is not designed for “storage array
> failover”. The array failover could be handled in the level higher than the
> DM-MP or the application level.  Secondly, the replication, most of times,
> is for disaster recovery between two sites. I would be interested in your
> use case why you would like to have array failover on the same Linux host.
>
Well, I've four nodes on cluster accessing the primary storage device, 2
nodes on build-1 and 2 nodes on build-2.
Something like below square at this image:
http://img139.imageshack.us/img139/642/clusters.jpg
If I can configure DM-MP to handle those two paths, in case of a "disaster"
(or even a storage failure, burning, crash...) i can easily promote the
another one to primary and all my systems (mainly use of cluster is a
file-server) becomes online again. I've no much to concern about data as
this is a sync replication (worst case if i got a filesystem/data corrupt).
Please, excuse me if i'm doing something nasty. This is my first work with
that kind of hardware and I have no friends to ask to.
And if you have any tips, hints, comments, recommendations for me, i'll be
really grateful :).
>
>
> Thanks,
>
> Yanling
>
>
>
Thank you,
--
Flávio do Carmo Júnior
Florianopolis/SC, Brasil
>   ------------------------------
>
> *From:* owner-t10 at t10.org [mailto:owner-t10 at t10.org] *On Behalf Of *Flavio
> Junior
> *Sent:* Friday, May 08, 2009 1:42 PM
> *To:* t10 at t10.org
> *Subject:* Re: WRITE_AND_VERIFY doubt
>
>
>
> Hi Konrad, thanks for your answer, but...
>
> I've already configured multipath to group my 4 paths to same LUN as a
> single device, the problem here is about 2 different LUN's.
> These LUN's are different because it comes from distinct storages devices,
> I want to group both because the contents is being replicated by DS4700
> Enhanced Remote Mirror feature.
>
> --
>
> Flávio do Carmo Júnior
> Florianópolis/SC, Brazil
>
> On Fri, May 8, 2009 at 1:50 PM, Konrad Rzeszutek <konrad at virtualiron.com>
> wrote:
>
> On Fri, May 08, 2009 at 12:52:28PM -0300, Flavio Junior wrote:
> > Hi folks, afternoon.
> >
> > I've a tad doubt when writing a simple code to sends a WRITE_AND_VERIFY
> > (0x2E) command to a scsi device.
> >
> > I need to do it to decide if this is a primary (and read-write) storage
> or a
> > secundary (and read-only) storage (ibm ds4700), and adjust my I/O
> MultiPath
>
> I was under the impression that the DS4700 was an RDAC device. Meaning
> you can use these settings:
>
>  device {
>	 vendor 	 "IBM"
>	 product	 "1814"
>	 hardware_handler    "1 rdac"
>	 path_grouping_policy	 group_by_prio
>	 prio		 "rdac"
>	 failback	 immediate
>	 path_checker	     rdac
>    }
>
>
> And it will detect whether your secondary paths are not in usage and
> call them 'ghost'.
>
>
> > priority accordingly, and i need to run it periodically to check if the
> > roles change.
> >
> > My doubt is, is there a area that I cant safety use this command without
> > risk to corrupt the data on disk? I'm not a really developer, just make
> some
>
> You will cause corruption.
>
> > researches and read some codes to assembly my own, here is the code:
> > http://pastebin.com/m22b06f04
> >
> >
> > O course, if there is a way more elegant to get this info I'll appreciate
> if
> > someone says me. :)
>
> I think using the right multipath.conf entry should do it.
>
> >
> >
> > Really thanks in advance and some sorries for english and dumb question.
> >
> >
> > --
> >
> > Flávio do Carmo Jùnior
> > Florianopolis/SC, Brazil
>
>
>



More information about the T10 mailing list