Re: AW: Network Storage Redundancy

I have 4 that would be in the cluster, load balancing apache and samba.

On Thu, 2002-04-25 at 00:39, Martin Bene wrote:
> Hi Matt,
> 
> > what are my options with failsafe for network replication of 
> > storage, as in the assosiated nodes have their own disks not sharing a 
> > single disk. Like GFS for example
> 
> I've been using drbd with heartbeat for quite some time now; recently I've also used it in a failsafe cluster - actually works quite well.
> 
> drbd gives you a block device that's replicated over the network, and (like shared scsi) may only be accessed on one node at a time. One downside is a 2-node limit, if that's no problem for you it's well worth considering.
> 
> Some hints on the use of drbd:
>  
>  - source is available at http://www.linbit.com/en/drbd/
>  - at the moment, the version to use is 0.6.1-pre10. 
>  - prefered kernel is 2.4.x, 2.2 could work (haven't tested with recent versions though)
>  - current version does NOT work with kernels with highmem support turned on, you MUST use a kernel compiled with highmem support turned OFF, so you're limited to 896MB Mem at the moment. This isn't a design limitation, just a bug in the current version.
>  - the drbd support in failsafe needs minor modifications in the resource scripts to work with current versions of drbd, see attached patch.
>  - system startup: you must start the drbd service before starting the fs_cluster service. drbd service startup ensures that data on both nodes is in sync before continuing.
> 
> Within the limits specified above I found drbd to be a quite reliable and usable solution for low-cost clustering.
> 
> Bye, Martin
>