[Linux-cluster] Clustering RHEL 2.1 without Shared Storage?

[Riaan van Niekerk]

Kit Gerrits wrote:
> Hey All,
> 
> I'm very new here, so this question may have passed by a few times already.
> I simply couldn't find a way to make it work, so now I'm asking the
> Professionals:
> 
> --
> Is there a way to run a 2-node RHEL 2.1 Cluster without the use of a Shared
> Storage box?
> The Quorum needs to be on any generic shared Block Device, but I'm not sure
> a NFS mount qualifies.


hi Kit

No, it does not. RHEL 2.1. clustering uses Raw devices/partitions for 
its quorum device.

> --
> 
> I can, if need be, mount a NFS share from either a Linux host or a NAS.
> 
> 
> Why 2.1?
> Because the application has only been certified for 2.1 and I'm too afraid
> of putting it on a machine with a 2.6 kernel.
> 

RHEL 3 and 4 contain the compat libraries for applications certified 
against RHEL 2.1 . If your application does not interface with the OS at 
a kernel level (or is coupled with custom hardware which only has 2.4 
kernel support), you may just "get lucky".

Unfortunately, RHEL 3 also still requires a shared device for quorum. 
Only in RHEL 4 does it become optional (although, in  my experience, a 
quorum device does wonders for the stability of a 2-node cluster)

> But, for the love of [insert deity here], WHY???
> Because we're already working on replacing the entire system, but this will
> take about a year and the current servers might die any minute now.
> 
> But if it's that important, why did you only just start the replacement?
> a/ I just got hired for the job
> b/ It's for a Government Organization, they're not that quick.
> 
> What's the current hardware?
> The main system is on HP/UX on HP9000/K360 (with a HASS and an AUtoRAID)
> There is a subsystem that has been tried and tested on RHEL2.1 on CPQ
> DL380-G3 (With a MSW500)
> I have 2 CPQ DL320 machines I can use for the back-up plan. (with no Storage
> Box)
> 
> 
> Can anyone please help me out here?
> 
> 
> Thanks,
> 
> Kit Gerrits

talk about being between a rock and a hard place. What you want, cannot 
be done on RHEL 2.1 without shared (block) storage. Give the customer 
the options and have them decide:

a) RHEL 2.1, without clustering (I would recommend this) but some kind 
of hot/cold standby mechanism (scripts) to periodically sync data to the 
standby host. If the primary host goes down fatally, have a process to 
manually switch over (this could imply starting the application, 
bringing up a virtual IP, etc)
b) RHEL 4 with clustering - breaks application certification, even 
though it might still work.
c) RHEL 2.1 with clustering (which requires the additional expense of a 
shared storage (fibre, not SCSI) storage array.

HTH
Riaan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: riaan.vcf
Type: text/x-vcard
Size: 310 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20061130/3d3f07af/attachment.vcf>