As a former analyst, Alessandro Perilli has found a way to see through the promises of cloud to the real-world problems every enterprise encounters when building a private cloud.
When we think about cloud, he says, we have certain expectations:
- Self service
- Standardize and fully automated
- Elastic (application level)
- Resource consumption-based
- Infinitely scalable (infrastructure level)
- Minimally governed
But what we actually get is dramatically different. (If you need a visual, Perilli compared our current expectations of private cloud to a winged Pegasus, and our actual progress on private cloud to a donkey to drive the point home.)
The differences between private enterprise clouds and public are enormous. Compare the list below to our expectations, above:
- The (enterprise) private cloud
- Undeniably complex
- Self-service (The only similarity to the above list)
- Partially standardized and barely automated
- All but elastic
- Mostly unmetered
- Capacity constrained
- Heavily governed
As an ex-Gartner analyst, Perilli shared what he called a (very mature) private cloud reference architecture. It includes many layers (5) plus self-service provisioning, chargeback, and self-service provisioning on top of a physical infrastructure. Several vendors (including Red Hat) can sell all of these components.
So what actually makes up a cloud? Do you need all those layers? And if you don’t have them today, when will you get them?
Gartner identified 5 phases to get you from a traditional datacenter to a public cloud. Getting from phase 1 (Technologically proficient) to stage 2 (operationally ready) can take years. That’s why Perilli recommends you take a “dev & test” approach to private clouds. “Very few private clouds are in production. There are a few industries where we’ve seen them successful.”
PARTIALLY STANDARDIZED AND BARELY AUTOMATED
“It’s easier said than done” to be fully standardized and automated. There’s a progression of complexity as you try to provision applications. The more components an app has, the more provisioning is required. The more pieces you try to automate, the harder your job will be. You also have legacy systems that need to be integrated. And when you design applications in the physical world, provisioning isn’t always well documented and, therefore, not repeatable. And, truth be told, it may take you 1 year to convert 10 applications (of the thousands in the enterprise).
ALL BUT ELASTIC
What does “cloud application” mean anyway, Perilli asked. The application should have cloud characteristics like rapid scaling and elasticity, self-service, etc. And it should be architected to be automated, failure aware, parallelizable, etc. And, Perilli, noted that most developers and applications don’t fit the cloud model. He says it’s a matter of culture and training. And most organizations, except for Netflix, don’t have everything it takes to do it right. “Enterprises take a long, long, long time to get there,” Perilli said.
Because we’re too busy with provisioning and orchestration issues, charge-back capabilities are taking a back seat. There are resources and licenses to manage, and if you aren’t careful, you’ll miss the charge-back opportunity.
There’s a massive need for governance, but customers aren’t successful in production clouds. You need an approval workflow, and there’s a need for applications to be removed when not needed. We need process governance, but aren’t at the maturity level to provide it.
So what do we do? Stay away from private cloud altogether? Nope.
ENTERPRISES SEE ENORMOUS VALUE IN PRIVATE CLOUDS
Speed to market and quality. Huge demand for Platform-as-a-Service (PaaS) bundles. Perilli says customers have shared these kinds of successes with him. The key, according to Perilli, is to buy only the cloud you need. Some providers sell you more than you need, and you end up using only 1/12th of their modules.
More tips from Perilli:
- Don’t believe the promise of fully automated production cloud. Build a test cloud instead and be pragmatic.
- Introduce support for scale out application in a meaningful way. Consider a multi-tier cloud architecture.
As a new Red Hatter, Perilli provided a quick overview of the capabilities of Red Hat’s cloud architecture. In particular, he pointed out CloudForms as a way to manage disparate systems under a single entity. And instead of buying more than you need, Red Hat plans to rely on certified ISVs to provide more management capabilities on top of our cloud. It’s a lean, smart approach, and one we’ll hear more of in the months to come.
Event: Red Hat Summit 2014 Date: 2:30 p.m., Wed April 16, 2014 Type: Session Track: Red Hat Cloud Technical Difficulty: 1 Title: Building enterprise clouds: Resetting expectations to get them right Speaker: Alessandro Perilli