High Availability for a University Website with Limited Hardware

I suggest using a Linux base operating system (OS) instead of Windows. Personally I like Ubuntu server because of the large and friendly community around it. Another good reason for using a Linux base OS is that it's easier to get free support. With Microsoft you often end up having to pay for support. Plus Linux is well known for having better high availability and security.

Thanks a lot for all your comments.

Greg, you advised setting up a Ganeti cluster. I researched this a bit and it certainly looks interesting but I'm afraid it would be too much of a hassle especially for someone who is unfamiliar with Linux.

I guess we would go for a RAID1 setup as choloepus suggested. You're right this is not exactly HA but lets hope it will suffice for now.

Could you please point me somewhere I could read about setting up RAID1 for this system?

Thanks a lot,
Nimi.

You might begin by defining what you mean by "high availability." RAID is for avoiding hardware failures. Clustering can protect against both hardware failures and system load. APC, Memcache and Boost are for performance/capacity enhancement, and while high system load can have the effect of an outage, it's not the same thing. Before you try solutions, you need to be clear about the problem.

I have a fairly low-powered Linux box running Drupal at Server Beach that's been rebooted maybe half a dozen times in five years, serving 500K to a million page requests/month. It's more likely that the power goes out in the entire region than the server has trouble. And there's nothing special about it at all. Runs APC, but not Memcache or Boost.

$ uptime
15:52:28 up 210 days, 14:18, 1 user, load average: 0.24, 0.39, 0.48

We're going to retire it next month, but I don't want it to know, lest it feel threatened.

Hi again,

I have succeeded to arrange another server so now we actually have:
2 servers with 4GB of Ram each
2 ISCSI HDs

I understand that there is no other option but to move to Linux to have a modest HA/Load-Balancing solution.

I've researched about Ganeti clustering gchaix recommended. Sounds very interesting. I think we will go for that.

The confiuration gchaix recommended is:

"4 virtual machines - 2 apache/php nodes (with APC), 1 mysql/nfs node, 1 varnish cache node. Set up the apache nodes to mount a shared webroot on the nfs node. Set up varnish to do a round-robin load balance on the apache nodes."

Just to be clear, is this the correct configuration?

Server 1:
1 apache/php (with APC)
1 mysql/nfs

Server 2:
1 apache/php (with APC)
1 varnish cache

Would anyone recommend following these instruction for configuring such a setup?
http://www.howtoforge.com/xen-cluster-management-with-ganeti-on-debian-l...

Many thanks,
Nimi.

You're close ... :-)

Set up the two physical machines as a Ganeti cluster. Then provision four virtual machines on that cluster: two webnodes, one MySQL, and one varnish. With virtualization like this, the two physical servers essentially become a single pool of server resources that the virtual machines draw from. Once the cluster is set up, you're essentially looking at a single system that hosts the virtual machines - the individual hardware that comprises the cluster becomes mostly irrelevant. Ganeti will handle the provisioning and decide which physical machine will be the primary node for each virtual machine.

Here's an overview of our Ganeti infrastructure: https://dokuwiki.osuosl.org/public/ganeti_osl_infrastructure
Here's a document on how we build Ganeti clusters on Gentoo: https://dokuwiki.osuosl.org/public/ganeti_cluster_gentoo
Here's some more on the basics of Ganeti: https://dokuwiki.osuosl.org/public/ganeti_how_it_works and https://dokuwiki.osuosl.org/public/ganeti_common_commands

Note - we use KVM instead of Xen for the virtualization. Ganeti supports both, but KVM is supported in the mainline Linux kernel, while Xen requires a custom kernel. For simplicity's sake, I'd go with KVM/qemu.