RES: [DRBD-user] DRBD + GFS + iSCSI?
leonardo rodrigues de mello
leonardo.mello at planejamento.gov.br
Tue Oct 17 21:13:29 CEST 2006
-----BEGIN PGP SIGNED MESSAGE-----
In the configuration i sent to the list,
The configuration is this:
(SOME STORAGE DEVICE) --------------- ( MANY CLIENTS)
To implement this some storage device i have sugested to use two
machines with drbd0.7+heartbeat+iscsi target
Only one drbd server will export the device using iscsi at one time. if
this servers fails, the other will become primary and start to export
the drbd device using iscsi to others machines.
This way, you will have one virtual san device with high availability
and fault tolerance.
Why to use iscsi on top of drbd, simple, drbd only work in two machines
environment, of course there is drbd+ for three machines environment.
But what i do if i need more than three machines ? i have to use other
way to export the device. (GNBD, AOE, ISCSI, ENBD, and others...)
another question is why i have to use drbd ?
I have to use drbd to implement failure tolerant and HA for the device
that will be exported.
And why i need to use a cluster filesystem?
Because if you have more than one kernel instance accessing the same
filesystem, you have to manage global lock, and others things that only
one Parallel or Shared filesystem can DO.
You can use one "normal" filesystem like ext3 for example, but that will
cause data lost and corruption.
Leonardo Rodrigues de Mello
CA Lists escreveu:
> My concern was that having multiple clients writing data to the same disk
> might be an issue for DRBD, but, based on your response, I'm assuming that
> as long as the filesystem support write ordering, DRBD doesn't really care.
> Is that an accurate assumption? Anyone else on the list done this before?
>> Yes it is possible, and there is people that had done something like that
>> You will do all that is done in the server part here:
>> forget about the windows client part in the document.
>> After you have one iscsi target under heartbeat exporting the drbd disk, you
>> will have to install the iscsi initiator in the clients that will access the
>> disk, and after that, install GFS in all the clients, pluss formatting the
>> iscsi exported disk with gfs.
>> You will use only drbd active/passive, there is no need to use active/active
>> in this situation.
>> You can use ocfs2 or any shared filesystem instead of gfs if you prefer.
>> Please, documentate this configuration and send to me if possible.
>> I had done this before, but didnt have time to documentate it yet.
>> Best Regards.
>> Leonardo Rodrigues de Mello
>> -----Mensagem original-----
>> De: drbd-user-bounces at lists.linbit.com em nome de CA Lists
>> Enviada: dom 15/10/2006 13:39
>> Para: drbd-user at lists.linbit.com
>> Assunto: [DRBD-user] DRBD + GFS + iSCSI?
>> I'm wanting to create a cheap, yet redundant, iSCSI SAN solution. My thought
>> was to get two separate servers, configure both with Linux-HA for failover,
>> iSCSI Enterprise Target on a GFS (or OCFS2) disk, then mirror the disk from
>> server1 to server2 via DRBD. I will need multiple clients (apache cluster &
>> mysql cluster) to connect to whatever server is active at the time. I know
>> that the current stable version of DRBD only supports 1 active system at a
>> time, which is all I'm going for, but I wasn't sure if it would allow the
>> disk to be exported via iSCSI and mounted on multiple clients at the same
>> time, or if it considered that the same as trying to mount it on both
>> servers at the same time.
>> In the end, my goal is to have an iSCSI target machine, with a second one
>> sitting there waiting (mirrored), just in case something goes wrong...
>> Is this possible? Thanks,
>> drbd-user mailing list
>> drbd-user at lists.linbit.com
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v188.8.131.52 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
More information about the drbd-user