The seven sins of private cloud adoption

Cloud service provider recommends regifting, getting close to users, and frequent capacity planning, but warns over unicorns
Written by Rob O'Neill, Contributor

The exhibition space at OpenStack’s Hong Kong Summit speaks volumes about the open source cloud platform’s quick progress since inception three and a half years ago.

Exhibitors include a few big names, the likes of OpenStack co-founder Rackspace, Redhat and IBM, but also many other companies that have seen the opportunity and launched the OpenStack ecosystem as start-ups.

Companies in the community now number over 200.

One of those is Metacloud, which provides OpenStack-based software called Carbon/OS and services to help ensure 24x7 private cloud performance.

Founder Steve Curry said shared his "seven deadly sins" of private cloud adoption with conference-goers this week.

1. Skipping up-front total cost of ownership analysis
Curry concedes private cloud TCO isn’t easy. Factors such as increased team and user efficiency, for instance, should be included.

A great question to ask while going through such an analysis is “What are you building for?” he said. Many do so to deal with peaks and troughs in demand which as start-ups grow can become costly on public clouds, prompting a private cloud project.

Curry said US$100,000 a month being paid to public cloud providers remained a magic number that would attract a “please explain” from the CFO.

Provocatively given Amazon’s recent price cuts, Curry claimed every time a public cloud provider reduces its price, it makes more money. That is because the cost of providing computing is dropping faster, in line with Moore’s Law, than the price reductions.

“If you have a private cloud, you pocket that value,” he said.

2. Forgetting the end user

Organisations should ask who they are building private clouds for, Curry said.

What they should not do is build based on their conception of what a private cloud should be. That will lead to user pain and frustration and drive “shadow IT” as users solve their problems for themselves.

Constant contact with end users is required.

“Needs analysis should never end. Make it easy to use and consume and make it reliable,” he said.

3. Mistaking the beginning for the end

“It’s not about the start, Curry said, it’s about the finish. Private clouds have a lifecycle including  capacity planning, monitoring, patches and major upgrades and have to be integrated with internal teams and policies.

4. Failing to consider new layers in IT

It is necessary to ensure existing internal teams can focus on their core responsibilities as clouds require unique engineering skillsets, Curry said.

Organisations building private clouds can choose to hire a cloud team, “capture a unicorn” or bring in outside assistance to manage their new platform.

Unicorns are multi-skilled people with deep engineering backgrounds and experience in running computer systems at scale.

“Amazon has burned down the only unicorn factory,” Curry said.

Such engineering skills are becoming rare and the few unicorns left are hard to keep because they want to be where the action is.

5. Overlooking the value of existing infrastructure

Curry said organisations adopting building private clouds should “regift” existing hardware such as servers, which are often only running at 10% to 15% capacity anyway.

Most, however, go out and buy new kit when that may be unnecessary. By using existing machines, the user is already starting to pocket the value created by Moore’s Law.

“OpenStack can run on almost anything,” he said.

6. Inadequate marketing

Users have to know the solution exists. IT needs to be transparent with them and show them the cloud roadmap just as a public cloud provider would with its customers, Curry said.

Users need to be educated, be part of the conversation and be connected to the private cloud innovation cycle.

7. Delayed capacity planning

Capacity planning needs to start straight away as degraded performance is a great way to get rid of users. 

Editorial standards