What’s with the private clouds?

In my previous post I discussed the fact that Clouds with all their fuzziness have come to mean self-servicing. But the form of self-servicing and what kind of service you are actually getting, is harder to grasp.

I consider the fuzziest of them all to be Private Clouds.

When they arrived I must admit that I could not see the difference between this and what we already had with virtualization in datacentres. Microsoft threw themselves on the bandwagon as well, and I wondered whether they were already backpedalling from the Public Cloud, seeing that they were not getting the business they had hoped for. I was seriously disappointed because the word “Cloud” was becoming so diluted, that I was afraid the whole idea behind it was evaporating into thin air. No cloud left.

It wasn’t until recently, attending a Windows Azure Summit in Seattle that it became clear – clouds are about self-service! A self-servicing data centre, locally? Or maybe remotely… but what would then “Private” mean? Let’s start off with looking at the reasons why companies are still keeping their data close at hand, or on premises as the term goes.

There are certainly very good reasons for this:

  • Security (Full control, stored locally, but maybe a bit like keeping your data in the mattress?)
  • Latency (The time it takes for the data to travel, physics still matter!)
  • Performance (The achieved data speed)
  • Up-time (Internet outage does not bring the company to a halt)
  • Law (Storing sensitive data outside your own realm often requires adhering to special laws and regulations)

I will not enter into a discussion on why companies prefer local computing here, but some of the reasons are subject to discussion. The important part is that these arguments are perceived to be valid, therefore much data will be kept locally in the foreseeable future. Only a very few medium sized companies are seriously considering getting rid of their own datacentre. They actually do exist, though.

The struggle of the IT-departments

Local datacentres must be controlled, because local servers are finite. Although the concept of virtual machines has helped to improve utilisation of physical machines, the physical machines are concrete devices that have physical limitations.

Seen from an IT-department’s view, users are notorious gobblers of memory and storage, forgetting to delete unused resources. Developers are obnoxious creators of “stuff” and they leave behind a mess of virtual machines, test machines and databases that may or may not be used. Only shutting down machines and removing them triggers a response, and then mostly a delayed one. Like 6 months later when a modification must be tested on a suddenly non-existent development environment.

Efficient management of internal computing resources is therefore only common sense.

So what about this private cloud thing? If we agree that “cloud” is about self-service, maybe access to internal resources could also be self-serviced? Because that is what the Private Cloud should be about. Imagine that you could create a virtual machine on the local datacentre, without asking anyone’s permission. “I need a database-server” can be translated into actually having one within 5 minutes. No other people involved.

What we need are self-service policies. Self-service policies to be implemented with data centre tools, so that IT transforms from being a gatekeeper, and so by definition a bottleneck, to a surveyor of the health of the data centre. IT should concern itself with limiting user quotas, and create templates for currently available virtual machines.

The difference in notion is important, but you might say that this is only a slight change of attitude. Maybe so, but new datacentre software has provided IT with the tools to work in this way.

A slight variation on the theme “Private Cloud” is the fact that some of the big hosting providers are boasting to be able to create a private cloud, just for you. This should mean that a finite number of servers are available at your disposal at their datacentre to play with, but the cost and investment would still be yours. I believe this makes it easier to extend your own infrastructure, and certain aspects of safety and security may be better taken care of than having everything in your own little data-room. However, it is still inflexible, and that means investments have to be made up-front, and adding new servers to the setup is probably as manual as it always has been.

Unless you share the server park with others. But that would make it a public cloud. So the tools for self-servicing are improving, but there is one important piece missing. Enabling infinite resources!