Yes. I've googled it.

I'm not satisfied that I understand clusters, at all, and I was hoping that someone like a LinuxQuestions Carl Sagan could answer my question...

Can a Cluster, made up of 5 Dell Poweredge computers, each with a NIC and a HDD, be a file-server that is highly available? If I go over, and break one of the computers with a sledgehammer, will the cluster continue to operate without data loss?

How does that work?

http://www.redhat.com/cluster_suite/

Goes on to say how fancy the cluster is, and some of the things that you can do with it, but does not address my question.

Hi,

What you are describing is what cluster are invented for. You can have for example the following basic setup, simplified to two nodes but you'll get the picture:
Set up two nodes (servers) with two disks in a raid1 configuration.
Created a LVM on the RAID1
Setup DRBD on both nodes
Install Apache
Install Corosync / Pacemaker on both nodes
Configure active / passive using Pacemaker and set it up so that the virtual IP, Apache and DRBD always stay together on the same host.

If you take one of the hosts down (cutting the power or hitting it frequently with your sledgehammer), the other one takes over. Of course in this situation with two nodes, if you kill one you no longer have a cluster. There's a lot of material on the internet about the things I mentioned here. You can even set up a cluster with DRBD over the internet to different physical locations so that if your building in Europe goes up in flames, your office in the US takes over. Lots of possibilities. If you want to try some out you can easily set those things up with VirtualBox, Proxmox, VMWare, ...

I've set up several clusters already with the tools mentioned above and with RHEL Cluster in a global networkk and if configured correctly you have a very high availability for your services. If you have more specific questions, don't hesitate to post them.

Kind regards,

Eric

1 members found this post helpful.

Thanks for your reply. DRBD is very interesting. Limited to a 2 node setup though, as you mentioned.

I'm interested in using the RHCS(centos version, really) to make a small cluster and demo it. But I'm stuck on this storage thing. DRBD is, very managerial overview here: rsyncing, the data from one machine to another and using a floating IP address. Really, we could *almost* do that with a script, no?

What gets me, as I was mentioning, is the storage in the cluster. With DRBD, you can have a drive on each machine, one goes down, the other comes up, and has the same data. How will a 5 node cluster, each machine having 1 disk (logically), accomplish this? Do they send data back and forth to each other as if they were each a disk in a raid array? Then if one goes down, the other 4 keep going?

Additionally, To test this, I was thinking of:

Make 5 VM's
Install Centos
Install Cluster Suite
( ? ) [some sort of howto for making a clustered fileserver]
Access the clustered fileshare
Turn off a VM
Access the clustered fileshare to verify it is still working

Hi,

DRBD is not limited to two nodes, I was only giving an example. You can have multiple disks on your servers in multiple locations. You can create different DRBD resources on different LVM volumes which overcomes the two-node limit for a part. You can even have a third node for disaster recovery included in the cluster. I haven't tried a DRBD setup across more then three nodes, but when I have some time (I already have 4 node cluster with everything but DRBD so shouldn't take to much time) I'll see if it really is limited to 2+1.
From the DRBD site
I advice you to read the DRBD site thoroughly. It mentions the advantages, disadvantages and possible work-around technologies to overcome disadvantages.

I've only worked with RHEL Cluster for Apache services not for storage so I can not really testify on RHEL Cluster. On Debian I've used a similar setup I described in my first answer various times. I also have configured the same setups using CentOS 5.5, using Pacemaker, DRBD and LVM and it works.

If you have the time you could setup one environment with RHEL Cluster Suite and another one with Corosync, Pacemaker, LVM and DRBD. I'm sure you'll find the use of Pacemaker very easy and a lot faster then RHEL Cluster Suite.

In your 5 machine setup, you could use three nodes for a filesystem, configuring one for disaster recovery and set up something like Apache on the other nodes, in active/active mode with load balancing and saving some small website in the DocumentRoot which you mount on the DRBD device. Then you can test some nice things in my opinion.

Kind regards,

Eric

2 members found this post helpful.