This article is more than 1 year old
Put your feet up and make the virtual data centre work for you
Control every layer - without upsetting your service provider
Many of the companies I have worked and consulted for over the years have rented server space from service providers.
You are only really likely to do that, though, if you are a relatively small company. Larger organisations are able to buy multi-tiered services so they control not just the configuration of the apps on the servers but every layer in the stack, right down to the storage.
So how does this work and how can you make it work for you?
On the face of it, the customer’s wish to manage every layer of the installation entirely contradicts the service provider’s desire to be able to control how the systems in its setup run.
The trick is for the service provider to provide at every level of the infrastructure, a secure “sandbox” within which customers can do whatever they wish without affecting anything outside it.
The core concept that we will refer to a lot is a virtual data centre. If you are a VMware user you will recognise this as a formal term used by vCloud Director, but I am using it here as a generic concept.
It is rather a good name, after all, as it sums up the concept of providing a customer with a virtual bundle of storage, servers, networking and applications.
Choose your provisions
Once you have a virtual data centre, you will be given an administrator account which has the ability to create other users’ IDs, with different privileges, within your own sandbox. So some users will be able just to open a console session, some to create and remove servers, some to stop and start servers, and so on.
With regard to provisioning the servers, let's start at the bottom with storage. Storage will always be presented via a SAN of some sort, and in the hypervisor layer of the virtual server the provider can define the amount of storage that is available to each customer.
While the most obvious configuration parameter is the amount of disk available, you also generally have the option of different classes of storage. It is common to see an expensive tier of high-speed storage, a modestly priced tier of mainstream storage and a cheap, very low-speed semi-offline tier.
This approach is enormously attractive to the customer because it means there is no need to over-provision servers. It is such a quick job to expand a virtual server disk that costs can be kept to a minimum and you can very easily assign different classes of storage to each application.
The only downside to using flexible storage is that it is not simple to reduce what you use. Most operating systems get very upset if you try to reduce the size of a volume because you can't guarantee that the bit you are throwing away doesn't hold data that you need.
So work around it. If you want to use some space temporarily, just define a new volume in your operating system and drop some storage on it. When you are finished with it, blow away the volume entirely. You can't shrink a volume but you can create and delete them at will.
Moving up to memory, you have no such restriction. Again the service provider can provision a RAM quota and you can assign it to your servers as you see fit. If you want to move it around between servers it isn't a problem (though you will probably have to live with a reboot).
Similarly with virtual CPUs, you can generally drop extra power in and then take it out without upsetting the operating system, though as with memory you will have to turn it off and back on again to action the change.
In the networking arena customers want freedom to do their own thing. The problem is that this could mean that one customer wants to use the same IP ranges as half a dozen others that live on the same platform, and eat more network bandwidth than the provider can shake a stick at.
Virtual firewalls are the answer. The provider configures a software-based firewall on the edge of the customer's virtual data centre, and any network traffic between that virtual data centre and anything outside it has to flow through that edge firewall.
If a customer chooses to have more than one virtual data centre within the hosted setup instead of one socking big virtual data centre with all its stuff in it, it will have to permit traffic between the two using firewall rules, or preferably set up a point-to-point VPN tunnel between the two.
That may sound unappealing but remember: if you are connecting into the hosted setup from your office you will want to set up a VPN at the edge of it anyway. It shouldn't be a great feat of technology to take the same approach between virtual data centres.
Standards, who cares?
When you are a customer of a virtualisation provider, you don't have to worry much about standardisation. You have no idea what the physical hardware is, you just know that your operating systems and virtual appliances need to support the virtual hardware presented to them by the hypervisor.
You don't really care about standardisation, then. What you do care about is abstraction – or, more accurately, consistency of abstraction.
Regardless of, say, the physical network cards in the hosts, what matters is that the particular virtual adaptor type configured into your servers is presented consistently by the infrastructure.
Similarly, you don't mind if the hardware has a dozen different CPU types as long as what is presented to you by the virtual hardware is the same across the board.