The rush for the (internal) cloud

With all the talk of clouds out there, I am seeing more and more companies making plans for internal clouds. I would argue however that internal clouds aren’t in fact ‘clouds’ as the industry currently defines them. Too...

Share

With all the talk of clouds out there, I am seeing more and more companies making plans for internal clouds.

I would argue however that internal clouds aren’t in fact ‘clouds’ as the industry currently defines them. Too often, there is a perception that you can purchase a cloud and install it. That perception points to cloud as a technology.

However, my view is that internal cloud is technology, organisation, process, and a financial model all rolled into one. In this post I’d like to talk about what a successful internal cloud, or as we’ve called it for the past 5 years at Accenture, Next Generation Data Centre, looks like.

First of all, buying some hardware and software and installing it in your data centre does not a cloud make. There’s no magic pixie dust that, when sprinkled on IT, will suddenly create an elastic set of resources that can dynamically be allocated to changing business needs.

Here, in no particular order is my checklist for a successful internal cloud. (As ‘internal cloud’ is what the industry uses, we’ll go with that terminology):

  1. A standard platform of open source, hypervisor and hardware. Commoditisation and standardisation at this level allow for underlying resources to be fungible
  2. An virtualisation stack that extends to networking and storage. Being fully virtualised allows for a wire-once infrastructure approach and once again buys flexibility and efficient use of resources
  3. Aggressive rationalisation of applications and the stacks they run on. Here at Accenture, we’ve standardised on one platform. This gives us consistency globally and the ability to support all regions. However, there is also the notion of using higher level application standards in the same vein as PaaS. A good starting point is to reduce the number of applications and homogenise as much as possible
  4. Aggressively move to automation for most standard IT service management tasks. You’ll always have some manual processes, but make them the exception
  5. Rethink the IT organisation. Has anyone been on a con-call with dozens of people, trying to remediate application problems? There’s too much balkanisation of skills and people in many IT organisations. Automation helps the folks in IT, too, and allows for more interesting work at the core, using the service model for internal IT as well
  6. Have the conversation with the business owners of applications. They need to buy in, too. By adopting a shared resource model, everyone can benefit. But having application owners who want bespoke hardware and who want nonstandard app architectures is a recipe for inefficiency
  7. Aggressively enforce the standards. Once you’ve gotten things cleaned up, automated, and virtualised do not allow untidiness to creep back in
  8. Accept the outliers. There will be legacy apps or infrastructure that simply can’t be made to run in the standard stack as the rest of the IT estate. Love them, cherish them, lock them away in a corner of the data centre and let them live out their lives in comfort
  9. Examine your enterprise cyclical rhythms. When are workloads running? Can you smooth out demand so that all your apps aren’t hitting peak at the same time? The idea is to maximise utilisation of your IT estate and not have too maintain pools of capacity to maintain peak performance
  10. Use public and external clouds for appropriate workloads. It’s probably not a differentiator for your enterprise to run a data centre. If it is, please by all means talk it up in your annual report. But, for everyone else I offer this question: If you want to buy a soft drink, do you necessarily want to own the infrastructure that provides it (the manufacturing plant, convenience store, refrigerator, etc) or do you just want the drink? There’s ways to get the results without owning the means to achieve those results

In all cases, of course your mileage may vary. But the lessons of the cloud providers should not be lost on those who feel it necessary to adopt an internal implementation of these concepts…and have fun while they’re at it!


Blog post by Joe Tobolski, Global Lead of Infrastructure at Accenture Technology Labs

"Recommended For You"

Would you like a data centre with that server? Crossrail contractor turns to Nutanix for automated cloud on the edge