Re: to cluster or not? what's best solution for 2-node HA?
On Jul 26, 1:50 am, Alex Samad <a...@samad.com.au> wrote:
> On Fri, Jul 25, 2008 at 06:09:24PM -0500, Ron Johnson wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > On 07/25/08 16:32, Bob wrote:
> > > Here's what I want:
> > > I have a vmware server running on etch, hosting 4 VM's.
> > > I want these 4 VM's to be hosted in such a way, that should anything
> > > fail on 1 VM host, there is another providing seamless service. I want
> > > to do this with the 2 pc's I already have, each with 2 NIC's, and no
> > > other hardware.
> > > My basic question - what's the best solution to accomplish this?
> > > Here's what I've looked at:
> > > drbd/HA (heartbeat) - allows you to raid1 a disk partition between 2
> > > pc's over ethernet
> > > issues: in the default primary/secondary failover config, if
> > > primary fails, there is some period of time (even if only seconds),
> > > required for the secondary to become primary - I can only assume this
> > > would mean rebooting all the VM's.
> > > however - you can run drbd in a primary/primary config - this
> > > sounds like what I want. But it sounds like I need a clustering files
> > > system to do this like GFS. After countless hours researching this,
> > > I'm still not sure how to do it - do I need GFS? OCFS? NBD?
> > > Now drbd isn't really a cluster, it's just raid1-ing 2 pc's - this
> > > could be all I need.
> > > But - would a REAL cluster be a better solution? I believe a cluster
> > > could provide load balancing, or at least optimized use of all
> > > available hard disk heads. Although, if drbd needs GFS, then in fact,
> > > doesn't this become a real cluster?
> > > What would a clustering solution look like?
> > > which is the best filesystem to use - GFS/OCFS ?
> > > would drbd still be needed if GFS is used?
> > > Any ideas, experiences, help - greatly appreciated!
> > A *real* cluster would entail running OpenVMS (since clustering is
> > built deep into the OS) on either HP Integrity servers or used
> > AlphaServers, and either buying little SAN boxes or using HBVS
> > (Host-Based Volume Shadowing, which is similar to llvm, but has been
> > in enterprise use for 25 years) on the disks.
> :) Tree hugger :)
> > With this, the OS will allow you to use both nodes concurrently on
> > the same data files, and in case of a node crash, the other node
> > will clean things up so that you don't have any corrupted data.
> > Next best would be Tru64 (a.k.a. OSF/1) Unix on AlphaServers,
> > because much of the clustering technology from VMS was ported to Tru64.
> Best to get some sort of shared storage, like a disk shelf that allow
> multiple connections - ie to server to connect at the same time (this
> can be done with scsi but usually with FC ), I believe esx has its on
> file system so just load it up, present the disk space to the 2 server
> and your off.
Alex - TX for your reply...
At the moment, I really don't want to invest in iscsi/FC tech... and
my solution doesn't really need the 99.999% uptime these solutions
might provide. I'd be happy with 99.0% ;-)
I do find your mention of ESX interesting though, as I have considered
upgrading my (free) vmware server to ESX. It does have some really
nice features. However, at the present point in time, that would add a
significant cost I'm not able to invest (I would need a couple add-ons
in addition to ESX as well).
So I'm really looking for the best open, linux based solution. The
more & more I look into this, the more it seems to me like drbd v8
running in primary/primary config, over GFS might be the best way to
I'm just curious if anyone disagrees, and if so, what their linux
based alternative might look like...
TIA - Bob