Tuesday 30 December 2008

Virtual Machines

So this is my first blog post and I thought I'd talk about what I've been looking into for HSL for 2009.

First on the list is virtualisation - for those that don't understand this, the basic principle is to take one physical computer (i.e. big metal box, or 'tin' as we call it (even thought it's not made of tin!)) and use some funky software to convert it into several 'virtual' machines. Each virtual machine looks and feels like a real computer (on the network at least), but is actually just a file on the server. They share the resources of the 'host' computer (disks, memory, processors etc.).

The obvious benefit is that you don't need to buy so many servers, which has various benefits (reduced overhead, lower running costs, greener IT as you aren't using excess power etc. etc.). However it doesn't necessary save on licensing costs (unless you are open source of course). Most servers are primarily idle so performance isn't a massive issue.

The good news is that the nice people at VMWare are currently giving away ESX Server for free - which lets you try out running virtual machines yourself for free. And it's a doddle to do.

We've used this kind of software before for infrastructure testing and software testing, but now we are venturing into running live servers using this technology - hosting Exchange Servers and providing standby recovery environments for our customers. None of this is live yet, but I'll keep you posted with our experiences into how this is going.

I have already used VMWare Server (latest version) to create a virtual Exchange Server for a customer which will be going live in early 2009, and plan to upgrade this to using ESX server in the future (to support running more virtual machines) as well as upgrading memory on the relevant hardware at the same time.

For those that don't know the difference, VMWare Server (or Microsoft Virtual Server) lets you run a virtual server 'inside' another server operating system (so you run Windows 2003 Server as normal, inside which runs VMWare Server, inside which runs another copy of Windows 2003 Server, Linux or whatever your favourite is). ESX Server is a 'hypervisor' which is basically a very small operating system that JUST runs the virtual machine environment, it's management software and not a lot else. This means there is more processing power, memory etc. left to run Virtual Servers, almost doubling the capacity to do this.

The benefits to us are that we don't have to buy a spare server for every customer we want to have a recovery setup for (which probably keeps them happy, as they'd be paying for it in one way or another), plus the setup time is much quicker (we can create a basic server config as a file, then just copy it around).

The other real benefit is portability - you can move a virtual server into totally different hardware and it still works. You can even do clever stuff with Storage Area Networks (SAN) or Direct Attached Storage (DAS) and higher end software that lets you move the virtual machines around whilst they are still running. This means we can start off recoving to an online VM and then download / transport the relevant files to a new set of hardware in a new office location.

We are using Dell PowerEdge 860 Servers with 8Gb RAM to handle this and are planning to implement a DAS/SAN behind this to handle the volumes of data and support cold/warm failover to similar hardware. This should allow us to run 4 VMs on each machine comfortably, potentially more if memory requirements are lower. We are only planning on running one or two 'live' servers with a few disaster recovery (DR) servers setup for customers.

Coupling this with an online backup solution will enable us to recover a customer to their last backup in a matter of minutes or hours rather than days or weeks, without having to invest heavily in additional licenses or in hardware - thereby keeping the costs down for our customers (and us!), whilst providing an excellent service. This forms a core part of our business continuity/disaster recovery strategy for both ourselves and our customers.

If anyone wants more detail about how we are planning to do this, or is interested in talking more about any of these elements please don't hesitate to get in touch.