Servers setup for middle-sized enterprise? specific-requirements inside [closed]

Question

I'd like to hear your opinions about how could be set-up servers architecture of the company which has about 10 infosystems with each of them having about 50 users concurrently and 2-3 infosystems of them have like twice per year for a day peak time of users up to 500 concurrently.
Users are not doing heavy-load operations, just plain HTTP GET and POST form submittings etc. The requierments would be that if one of the servers hardware like motherboard would give up then it should only take about up to an hour to have the server running again.
Other requirement would be that it should take not much time to transfer whole infosystem (files, database - total up to 50GB) from a production environment to test environment and vice-versa.

What could be reasonable architecture for that? I don't care if you tell concrete examples by vendors or just the different technologies. Maybe even some links with advices.
I'm not an expert in that and the architecture atm is each infosystem is having a standard computer box for itself.
For example HP Proliant DL380 G4 with Xeon 3.2GHz and 4 GB RAM. Other one is Tigma with 2*Xeon DC 1.6GHz and 4GB. Also some Xen virtualization is made and on that Tigma machine and has 2 infosystems running.

But the problem with these standard builds is that if the server gives up, for example motherboard, then it would take a day or two to setup the infosystem again on a new server - it would involve of having a spare empty server or virtualizing existing server.

As an other approach would be of having a SAN for persistent storage and 2-4 powerful servers connected by iSCSI (i could be totally wrong) with virtualized infosystems on them. That would mean if one of the servers burns out for example then it would take probably only some hours if not less to point that infosystem files and database in SAN to another server. Other good advantage of that is that advantage of the Xen dynamic resource sharing would give more resources for the infosystem when it has its peak time of users.
What are the disadvatages for that approach? If the SAN box breaks then all the persistent data would get lost. Should there be just a back-up SAN or another SAN with intense replication?

I'm sure that there are many other approaches and in my described ones i could be wrong with the architecture facts anyway. What is the most reasonable build-up for these requirements? The more expensive blade servers are meant for space saving and less cabling which pays itself where room rent is high but it's not an issue here. Blade servers are less in size so they demand more cooling and consume more energy which is less green-thinking that non-blade server...

Answer 1

I would suggest taking a look at VMWare, especially "Live Migration". The higher-end VMWare products allow you to run your virtual machines from a SAN, and move the virtual machine from one server to another pretty much on the fly. If you need more resources for a VM on certain days you can power up a dedicated machine for that VM for those days and migrate the VM to it. When the demand goes down, you can migrate it back to sharing with other VMs allowing you to turn off the dedicated machine, saving power and cost.

Also if the server dies, you just start the VM on another server in seconds.

Take a look at the ESG video on the right of the VMWare virtualization webpage - it's quite interesting.

As for SAN reliability - a SAN can be spread across multiple physical units - even multiple sites - with different levels of replication - live replication for locally connected SAN units, snapshots for remote ones, backups, etc. Just make sure you have a dedicated ultra-fast network for linking your local SANs (minimum if 2 Gigabit bonded channels ideally).

As far as hardware? Well, with VMWare it really doesn't matter too much, but a nice rack of 1U servers (Dell do some good low-cost ones) won't take up too much space. Don't forget to link them up with a good high-quality switch, or even a pair of switches, with redundancy so if one dies you don't lose your entire network.