On Thu, 7 Feb 2008, Atanas Dyulgerov wrote:
I'm not going to compare both solutions. I'm creating a cluster with 3 nodes -
two active and a passive. Two special applications are going to work on both
active nodes. If either of them fails it will be 'migrated' to the standby
node. Those applications use heavily the storage device. For that purpose the
storage device must be shared on all three machines. Each application uses its
own part of that device (different folder or different partition). They does
not access the same data.
Due to performance issues using NFS is not an option for me. What I need is
exactly a way to export/import a block device over the network. There are the
following alternatives for network block device:
- NBD - legacy, does not handle with the network failure
- BNBD - read only
- ENBD - works, but there are performance issues and again network failures are
not handled properly. Works well with pure kernel, but not with RH or SUSE
kernels. Not documented. Seems legacy as well. DM-multipath does not support
ENBD devices.
- GNBD - This seems the best of the above. Supports fencing and multipath. This
works well for me ...
... BUT GNBD requires rh cluster manager and works tightly with RH Cluster
Suite. Needs all RHCS packages installed. Does not work on most Linux
distributions. So, if I want to use GNBD (with whatever FS - ext3, gfs, ocfs2 -
it does not matter) I'm forced to use RHCS on every node.
What about iscsi? Would that work for you? Several of us are
working on hot failover heartbeat managed iscsi clusters as backends
to frontside application clusters.
I really like Heartbeat and got used to it but it is stupid to use
Heartbeat on top of RHCS. Better use RHCS only.
So my question is, is there an NBD solution (an GNBD alternative)
which works with Heartbeat? I couldn't find such. So no way for me
to share a block device over the network in my Heartbeat cluster.
I have to switch to RHCS...
Do you plan to implement/adopt such NBD and Global File System
which will be manageable in Heartbeat?
I am not sure if GFS and companions such as CLVM and DLM can be
managed by heartbeat. This is on the research list but I've not had
time to find it yet.
If GFS is not manageable by heartbeat (and I think GFS is the
cluster file system I would prefer to use) then the OCFS2 cluster
file system could be used; in its latest versions it apparently
supports DLM, though I've not used it that way.
-- Michael
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
