Gary Lamb remembers working on one cloud computing migration project so tricky that it was never completed. It’s an object lesson on how the seemingly boundless promise of the cloud can get mired in earthly concerns.

The government agency attempting the migration was behind the eight ball from the beginning, according to Lamb, chief technology officer at the information technology consulting firm Enterprise Networking Solutions Inc., which spearheaded the move. Fearing the loss of the agency’s business, the existing cloud provider had blocked access to the tools that would have enabled the agency’s roughly 30 virtual machines, or computers emulated in software, to be easily replicated to another cloud.

“So we fell back to backing up and restoring it,” said Lamb. The problem: The 25-terabyte workload was too big to move over a three-day weekend using available network bandwidth. Lamb and his team came up with the idea of using backup software to move the large base image — which doesn’t change — over a period of several days and then migrate the smaller daily updates called differential images each day until the cutover.

But the cloud provider’s backup software wouldn’t cooperate. It insisted on moving both the base and differential images prior to restoring on the new cloud service. That made the move an all-or-nothing proposition, and there was no choice but to abandon the project. “The customer never made the shift, and they weren’t happy,” sighs Lamb.

Welcome to the reality of multicloud, a world in which organizations can freely shift workloads and data among cloud providers to get the best combination of services, cost and performance – at least in theory. The nitty-gritty details of making such portability a reality are a lot tougher.

That isn’t dampening enthusiasm for the multicloud concept, which a full 81 percent of enterprises are currently pursuing, according to a recent survey by RightScale Inc. Research firms including Gartner Inc. and International Data Corp. have forecast that multicloud will be one of the dominant forces driving cloud growth over the next several years.

But frontline companies that are struggling to reduce the dependence of applications on underlying cloud platforms have encountered an assortment of difficulties ranging from cloud provider intransigence to the practical challenges of shifting large databases over unreliable internet connections.

Lamb ultimately discovered JetStream Software Inc., a startup that has developed a cross-cloud platform that enables workloads to be moved between cloud platforms in near-real-time. But the discovery came too late to save the business with the government agency. Customers often have to turn to third parties to solve problems because the cloud platform providers have little interest in making migrations easy.

It’s easy to underestimate how many more variables a multicloud environment can introduce to IT infrastructure management. “If you’re operating Amazon Web Services along with your own infrastructure, that’s complex enough,” said David Linthicum, chief cloud strategy officer at Deloitte Consulting LLP and a prominent writer and columnist about cloud-related topics. “Add in Google, Alibaba and managed services providers, and people are getting beyond their tipping point.”

Linthicum coined the term “cloud complexity management” to define what he expects will be a new discipline that will be needed to coordinate far-flung distributed infrastructure. “Once you get to 2,000 or 2,500 virtual points to manage, things start to break,” he said.

Multicloud by necessity

Most organizations already use several cloud applications from different providers, but when it comes to infrastructure they tend to stick with a single vendor. The term “multicloud” refers to an emerging IT architecture in which several infrastructure-as-a-service platforms are deployed, whether by design or by necessity.

In many cases, it’s the latter. “If you gave CIOs the choice, they would use only one cloud because it’s easy,” said Craig Lowery, a research director in the technology and service provider group at Gartner Inc.

But CIOs don’t necessarily have a choice. Mergers and acquisitions may bring additional cloud providers into the fold. Developers demand features available only in a cloud that’s not on the approved list. Or perhaps the decision is dictated by regulatory restrictions.

The eagerness of organizations to adopt cloud platforms without a coherent strategy has contributed to the confusion. In particular, the rush to adopt public cloud computing, as much for agility as for cost efficiency, resulted in a lack of focus on governance, according to Rhett Dillingham, a senior analyst specializing in cloud infrastructure services at Moor Insights & Strategy. “A lot of multicloud use is driven by independent decision-making by different teams within a company who have their own preferences or existing relationships,” he said.

There are good reasons organizations choose to use more than one cloud platform. They may want to take advantage of the strengths of certain providers, such as Google’s machine learning wizardry or Amazon’s vast scope. Having the ability to shift workloads among platforms gives customers pricing leverage, protects against outages and mitigates the risk of being locked into one supplier, a perennial concern of technology buyers.

And sometimes a multicloud approach is needed to have a shot at winning business in a region where processing and storage must be local, as is the case in many parts of Europe. “You bear the cost you have to bear to get that money,” said Sam Ramji, vice president of cloud platform at Autodesk Inc. The bottom line, Lowery added: “You’re going to be in multiple clouds whether you like it or not.”

Many choices, little consistency

There are no fewer than five big pain points for organizations attempting to manage multiple clouds, according to Dillingham: cost management, security and compliance, application development, deployment and operational management.

The problem is that each cloud provider has its own set of applications, services, interfaces and management tools, most of which are at least in part proprietary. Although everyone supports a common set of operating systems such as Linux and Windows, “the problems are above the operating system, things like how the network is configured and how load balancing works,” said Lowery. While the major cloud providers all offer object storage, for example, “Google’s object stores work differently from AWS’.”

That adds up to operational overhead for any customer that wants to expand its cloud options. For example, Autodesk has a staff of 80 IT professionals managing its AWS operations. “To add a second cloud I’d probably need 40 more,” Ramji said. “I’d need to show a lot of business value to make that case to my CEO.”

Ramji compared the current scenario to supporting a combination of Apple Inc. Macintoshes and Windows personal computers. “They’re both computers, but the tools you need to manage them are quite different,” he said.

CIOs shouldn’t expect the major cloud providers to provide much help resolving the complexity issue. Analysts said Amazon, which declined to be interviewed for this article, only just acknowledged the importance of hybrid cloud with the recent release of an on-premises offering that duplicates the cloud giant’s virtual infrastructure. It has acknowledged that customers want to use more than one cloud but has deferred to partner VMware Inc. to support them.

“Does Amazon want to help multicloud?” asked Stu Miniman, senior analyst at Wikibon, SiliconANGLE’s sister market research company. “Not right now.”

Microsoft has been more welcoming, even declaring last September that its future success depends upon supporting customers’ multicloud ambitions. The company acquired multicloud management startup Cloudyn Software Ltd. in mid-2017, and today markets the software as a cost-optimization service.

“Multicloud adoption has many challenges including technology, user and process issues,” a Microsoft spokesman said in an emailed statement. “Customers must develop teams of experts who believe in their respective cloud vendor. Organizational challenges can be tackled by leveraging programs, like the Microsoft partner ecosystem and developing internal organizational plans, to nurture talent for these cloud platforms.”

Of the big-three cloud providers, Google has embraced the multicloud agenda most fervently. “We give customers freedom of choice and we don’t lock them into our platform,” said Chen Goldberg, director of engineering at Google Cloud. “We want them to use our platform not because they’re locked in but because it’s the best cloud.”

Google’s position isn’t surprising in light of its third-place market position. It has the least leverage over customers and the most to gain by making it simple for them to shift workloads. The company has been the most proactive of the big three providers by support products such as Stackdriver for multicloud monitoring, the open-source Spinnaker for cross-cloud application development and its own Apigee for managing application program interfaces across multiple clouds.

Google has another agenda, too. It’s the principal developer of Kubernetes, the open-source container orchestration platform that promises to create a consistent operating layer across cloud platforms. With support from every major public and private cloud infrastructure-as-a-service supplier, Kubernetes can make software encapsulated in the small, portable virtual machines called containers run on any cloud platform running Kubernetes – at least in theory.

Google continues to be the top code contributor to the Kubernetes ecosystem and its greatest cheerleader. “We are all in on Kubernetes,” Goldberg said. “We want Kubernetes to be successful.” That appears to be a foregone conclusion: Wikibon predicts that within the next five years 90 percent of multicloud applications will use Kubernetes.

The K8s factor

The overwhelming success of Kubernetes (often abbreviated K8s as shorthand for the first and last letter along with the number of letters between) has helped to fuel growing interest in the multicloud model, but orchestrating containers only solves part of the problem.

In order for applications to be fully portable in a Kubernetes environment, they must be written to run entirely in containers, a fact that rules out most legacy software. Kubernetes is also dependent upon whatever local storage is available, which varies with the environment. And if applications take advantage of any native features of the host cloud, such as a proprietary database or development framework, those dependencies are lost when moved elsewhere.

Bala Rajaraman, an IBM Corp. Fellow who specializes in hybrid cloud, offers the example of a service that accesses a local database. “You need scale and want to move that service to a public cloud,” he said. “Now it has to go over a wide-area network to reach the database. It’s going to have horrid performance.” Moving applications between clouds has the same perils.

Kubernetes has been characterized as a multicloud panacea, but it’s really just a foundation for others to build upon, said Joe Beda, chief technology officer of Heptio Inc. and one of Kubernetes’ principal developers, in a recent interview on theCUBE, SiliconANGLE’s video platform.

“From the very start, we never pretended that Kubernetes was going to be this magic layer that was going to make differences between different environments disappear,” he said. “What we did want to do was find the commonalities and minimize the extra differences that didn’t need to be there.”

The situation is likely to improve over time as cloud-native commercial and open-source products fill in the gaps, but at the moment organizations must containerize every component of the application and its dependents to have a chance at enabling multicloud portability.

“The idea is that instead of depending on services within the clouds, I bring my services with me,” Lowery said. “It’s my own little city made of containers.” Providers are increasingly offering their products in pre-built containers with these services already embedded in them, he said, but that doesn’t obviate the need for developer discipline.

The market responds

With cloud giants showing only modest enthusiasm for portability, a host of outside companies is stepping up to the plate. Some major computing vendors are focusing on multicloud support after having failed to break into the top echelon of public cloud providers on their own. Established systems management providers such as CA Technologies Inc. and BMC Software Inc. are burnishing their portfolios with multicloud features and a small army of startups is charging in.

IBM, for example, is pushing multicloud management using a platform that relies heavily on containers. It has also positioned its planned $34 billion acquisition of Red Hat Inc. as a bid to become the leader in multicloud platforms. “If you have a fairly complex environment and if it doesn’t get integrated in reasonable way you aren’t going to have the agility that cloud provides,” Rajaraman said. IBM also continues to offer its own infrastructure as a service.

VMware Inc., whose Amazon partnership puts it in a unique position to reach some of the largest cloud adopters, updated most of its core software and storage virtualization platforms last summer to provide better support for multicloud deployments. It also acquired CloudHealth Technologies Inc., developer of a cloud operations platform that spans AWS, Microsoft Azure and Google Cloud.

Hewlett Packard Enterprise Co. is touting OneSphere as a way to deploy, scale and manage applications across multiple private and public clouds. Cisco Systems Inc. is targeting similar features with CloudCenter, a multicloud management tool that it will likely highlight at its Cisco Live conference in Barcelona starting Jan. 29. Other prominent established players with new multicloud management ambitious include Nutanix Inc., Mesosphere Inc., and Veritas Inc.

Does this mean the promise of fully portable workloads is on the horizon? Probably not, said Deloitte’s Linthicum. Providers are still focused more on extending their own products than integrating with each other.

“We’re leveraging all these new technologies, but the legacy stuff doesn’t go away,” he said. “Now we have these other entities that are ecosystems of their own. They’re building their own silos.”

And for many users, multicloud isn’t a big priority, anyway — at least not yet. Most organizations are likely to continue to run their critical applications on a single platform, with perhaps a second used as a backup for the sake of reliability and simplicity, said Milin Desai, vice president of cloud services at VMware.

“In a perfect world, customers say they don’t want to be too dependent, but typically most apps leverage one single cloud provider,” he said. “The idea that you’re going to run anywhere willy-nilly is really difficult to do.”

For organizations that want true intercloud portability, there are already solutions in place, but they require tradeoffs. For example, instead of using a cloud-branded database management system, companies can choose an open standard such as Postgres or MongoDB. Similarly, using Apache Kafka instead of a custom alternative such as Amazon Managed Streaming for Kafka enables an application to process streams in any cloud that supports Kafka.

But as anyone who has tried to build cross-platform applications on a personal computer or server will tell you, portability has its penalties in the form of reduced functionality and often slower performance. That’s nothing new to chief information officers, though. “Trading off portability for functionality is a law of the universe,” said Gartner’s Lowery.

Data gravity

But in the long run, users can expect to have more flexibility in the future to deploy their cloud workloads wherever they want them. That’s likely to shift the competitive stakes for cloud providers as well as the services they offer their customers. The defining question will become “Who controls the data?” said Peter Burris, Wikibon’s chief research officer and general manager.

The law of “data gravity” says that large bodies of data tend to attract applications, services and other data. That’s why there’s a land grab going on in the cloud for customer information, with the three biggest services offering 28 branded database management systems among them along with an assortment of storage services. The thinking is that the service that gets the lion’s share of a customer’s data will also get most of its processing business.

“Data gravity is what will keep customers in a particular cloud,” said VMware’s Desai. Added Autodesk’s Ramji, “Kubernetes doesn’t solve the data gravity problem at all.”

That fact is likely to set off a tug-of-war between cloud providers and their customers. Savvy IT managers will realize that shifting all their data to a particular cloud provider essentially binds them to that platform because of the time and cost of moving around large databases, as dramatized by the anecdote at the top of this story. “CIOs will want the control point inside their four walls,” Burris said.

Startups such as Scality Inc., Cloudian Inc., Caringo Inc. and SwiftStack Inc. are answering the call with cloud-independent storage.  They’re joined by stalwarts such as IBM and even records management giant Iron Mountain Inc. “There’s an emerging idea of a cloud volume” that’s equivalent to the storage areas with a single file system on a hard disk, Desai said. The downside: It’s yet another entity to manage.

And that points up the irony of cloud computing: It was initially cast as a simpler way to provision and manage infrastructure, but the profusion of platforms and lack of established practices for managing them has actually made the whole environment even more complex.

As the multicloud muddle likely gets worse before it gets better, IT organizations and providers alike have a lot of work ahead to make cloud computing fulfill its promise.

Photo: Brian Cook/Unsplash