Re: EVMS or Cluster LVM

Hi,

The use of standard LVM on data cluster is dangerous. A error can easily break the server and as we are many administors servers, the risk is too great. I could use LVM if I was sure that the SLES 11 will have CLVM in standard else it seems me dangerous.

My goal is to optimize the server load by moving the domU from one server to another or to do maintenance server with short stop (standard or live migration). Therefore, the LUN must be mounted on all servers at the same time. 

I also thought about setting up many partitions on the disk LUN but the problem is hot scan of the partition table changes as SCSI devices. On Sles 9, I need to reboot the server to use the partition table changes. I have not tried on the sles 10 sp2. With this solution, I lost snapshot functionnality too. May be that I can mount this partitions on another server in read only to make the file backup.

-----------------------
Nota : On OES installation documentation (1 or 2), I found this : "the Novell Storage Services? (NSS) file system requires EVMS as the volume manager of its device in order to use the full range of services that NSS offers. Therefore, EVMS must be the volume manager of any device where you plan to create NSS data volumes." 

Now in new cluster documentations or presentations since end 2007, Novell don't talk about EVMS anymore but only of OCFS2 and iSCSI. If EVMS is depreciated then what's the new Novell politics ?
-----------------------

-- 
Olivier Boyaval
Direction du Système d'Information Hospitalier
Responsable équipe maîtrise d'oeuvre informatique
Tél : 03.21.64.42.20
Fax : 03.21.64.44.38 Alexei Roudnev <[email protected]> a écrit :

> I use:
> - plain LVM
> - + heartbeat
> - + normal file system (reiserfs)
>
> and it works pretty well, after a few modifications to the LVM script (to
> make sure that when cluster switchover, LVM release disk groups).
>
> An idea is that you can use usual LVM, but cluster must maintain it
> mounted on the single host only. So I can modify LVM group without any
> problems and can switchover group to another server (and I can configure a
> few resources in heartbeat so that by defaulkt they spread over 2
> servers).
>
> I really DON'T RECOMMEND running any VM (Vmware or XEN, dont make any
> difference) on OCFSv2 (it's curious but OCFSv1 was suited for VM
> hosting much better) -  OCFSv2 makes too many resource locks and XEN (or
> VMWare) can't lock disk image on the single node when VM guest is alive,
> so it cause a huge extra load of the OCFSv2 and wil work very slow.
>
> You can try non-clustered LVM for the clustered data too, but (as was
> saying earlier) you must be sure, that when you change LVM layout, disk
> group is mounted (I mean vgscan groups) on the single node only.
>
> Alex Roudnev
> Exigen Group
> 415 806 3741
>
>
>
>
>
> "Ciro Iriarte" <[email protected]>
> Sent by: [email protected]
> 06/06/2008 05:02 PM
>
> To
> "Olivier BOYAVAL" <[email protected]>
> cc
> [email protected]
> Subject
> Re: [suse-sles-e] EVMS or Cluster LVM
>
>
>
>
>
>
> 2008/6/6 Olivier BOYAVAL <[email protected]>:
>>  Hi,
>>
>> I want to create a farm Xen server on a SAN. I don't want to work with
> disk
>> file for DomU and I would like to use LVM to manage partition on the
> LUN.
>>
>> LVM isn't cluster aware and Novell docs tell us to use EVMS instead.
>>
>> But EVMS is depreciated, no update since 2006 and IBM discontinued
>> development of this project.
>>
>> I found some web page on CLVM but SLES LVM2 isn't compiled with it.
>>
>> Any ideas to make this farm server other than using ocsf2 ?
>>
>> --
>> Olivier Boyaval
>> Direction du Système d'Information Hospitalier
>> Responsable équipe maîtrise d'oeuvre informatique
>> Tél : 03.21.64.42.20
>> Fax : 03.21.64.44.38
>
> I'm seeing i/o wait of about 100% with file disk images on OCFS2 and I
> agree it's not a good idea... I'm going with a LUN per domU as i'm not
> going to run too many of them, if you're the SAN admin or you can nag
> him frequently, it's a good route.
>
> Another option is to use plain LVM, the only catch is that when you
> need to modify metadata (resize, new volume creation), only one node
> can be online.
>
> Regards,
>
> Ciro
> _______________________________________________
> suse-sles-e mailing list
> [email protected]
> http://listx.novell.com/mailman/listinfo/suse-sles-e
>
>


----------------------------------------------------------------
Centre Hospitalier de Béthune

_______________________________________________
suse-sles-e mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-sles-e