Most enterprises use more than one of the hyperscale cloud providers, but "multicloud" remains a partitioned approach...
for corporate IT.
Amazon Web Services (AWS) continues to dominate the public cloud infrastructure market it essentially created a decade ago, but other platforms, especially Microsoft Azure, gained a foothold inside enterprises, too. As a result, companies must balance management of the disparate environments with questions of how deep to go on a single platform, all while the notion of connectivity of resources across clouds remains more theoretical than practical.
Similar to hybrid cloud before it, multicloud has an amorphous definition among IT pros as various stakeholders glom on to the latest buzzword to position themselves as relevant players. It has come to encompass everything from the use of multiple infrastructure as a service clouds, both public and private, to public IaaS alongside platform as a service (PaaS) and software as a service (SaaS).
The most common definition of a multicloud strategy, though, is the use of multiple public cloud IaaS providers. By this strictest definition, multicloud is already standard operations for most enterprises. Among AWS customers, 56% said they already use IaaS services from at least one other vendor, according to IDC.
"If you go into a large enterprise you're going to have different teams across the organization using different cloud platforms," said Jeff Cotten, president of Rackspace, based in Windcrest, Texas, which offers managed services for AWS and Azure. "It's not typically the same product teams leveraging both platforms. It's often different business units, with a different set of apps, likely different people and organizational constructs."
The use of multiple clouds is often foisted upon enterprises. Large corporations may opt for a second vendor when their preferred vendor has no presence in a particular market. Typically, however, platform proliferation is driven by lines-of-business that either procured services on their own or were brought under an IT umbrella through mergers and acquisitions.
"By the time these two get to know each other it's too late and they've gone too far down the path to make the change," said Deepak Mohan, research director at IDC.
Dave ColesanteCOO, Alert Logic
An apples-to-apples comparison of market share among the three biggest hyperscale IaaS providers -- AWS, Azure and Google Cloud Platform (GCP) -- is difficult to surmise because each company breaks out its revenues differently. Microsoft is closing the gap, while GCP saw a significant bump in 2016 as IT shops began testing the platform, according to 451 Research. But by virtually any metric, AWS continues to lead the market by a sizable margin that is unlikely to close any time soon.
Nevertheless, the competition between the big three is not always a fight for the same IT dollars, as each takes a slightly different tact to wooing customers. Amazon, though softening to hybrid cloud, continues its stand-alone, all-encompassing approach, while Microsoft has a greater percentage of enterprise accounts as it positions itself to accommodate existing customers' journey from on premises to the cloud. Google, meanwhile, is banking on its heritage around big data algorithms, containers and machine learning to get ahead of the next wave of cloud applications.
"[IT shops] are not evaluating the three hyperscale guys purely on if AWS is cheaper, or which has the better portal interface or the coolest features because there's parity there," said Melanie Posey, research vice president at 451. "It's not a typical horse race story."
The move away from commoditization has also shifted how enterprises prioritize portability. In the past, companies emphasized abstracting workloads to pit vendors against each other and get better deals, but over the past year they have come to prize speed, agility and flexibility over cost, said Kip Compton, vice president of Cisco's cloud platform and services organization.
"We're actually seeing CIOs and customers starting to view these clouds through the lens of, 'I'm going to put the workloads in the environment that's best for that workload' and 'I'm going to worry a lot less about portability and focus on velocity and speed and taking more advantage of a higher- level service that each of these clouds offer.'"
Silos within a multicloud strategy
Even as the hyperscale vendors attempt to differentiate, picking and choosing providers for specific needs typically creates complications and leads to a siloed approach, rather than integration across clouds.
"It's more trouble than it's worth if you're going to do it that way," Posey said. "What ends up happening is company XYZ is running some kind of database function on AWS, but they're running customer-facing websites on Azure and never the two shall meet."
The idea of multicloud grew conceptually out of the traditional server model where a company would pick between Hewlett Packard Enterprise (HPE) or IBM and build its applications on top, but as the cloud evolved it didn't follow that same path, Mohan said.
"The way clouds were evolving fundamentally differs and there wasn't consistency, so integrating was hard unless you did a substantial amount of investment to do integration," he said.
It is also important to understand what is meant by a multicloud strategy, whether an architecture supports a multicloud strategy or that workloads actually run on multiple clouds.
"There's a difference between being built for the cloud or built to run in the cloud, and it's difficult from a software development perspective to have an architecture that's cloud agnostic and can run in either one," said Dave Colesante, COO of Alert Logic, a cloud security provider in Houston.
Alert Logic is migrating from a mix of managed colocation and AWS to being fully in the cloud as it shifts to a microservices model. The company offers support for AWS and Azure, but all of the data management ends up back in AWS.
The company plans to design components of its SaaS application to provide flexibility and to assuage Microsoft customers that want the back end in Azure, but that creates limitations of what can be done on AWS.
"It's a Catch-22," Colesante said. "If you want to leverage the features and functions that Amazon makes available for you, you probably end up in a mode where you're hooked into some of the things."
The two key issues around multicloud center on the control plain and the data plain, IDC's Mohan said. A consistent way to manage, provision and monitor resources across all operational aspects of infrastructure is a challenge that's only exacerbated when enterprises go deeper on one platform than another.
On the data side, the concept of data gravity often discourages moving workloads between clouds because it's free to move data in, but expensive to move data out. There are also limitations on the speed and ease by which they can be migrated.
Getting the best of both worlds
Companies with fewer than 1,000 employees typically adopt a multicloud strategy to save money and to take advantage of new services as they become available, but the rationale changes with larger enterprises, Mohan said.
"As you move up the spectrum, the big reason is to avoid lock-in," he said. "We attribute that to the nature of apps that are being run, and that they're probably more business-critical IT app run by organizations internally."
The largest organizations, though, seem get the best of both worlds.
"Especially if it's for experimentation with new initiatives, they have much higher tolerance for going deep onto one platform," Mohan said. "For bread-and-butter workloads, volatility and jumping around services is not as important."
At the same time, large organizations that prioritize reliability, predictability, uptime and resiliency tend to favor the lowest common denominators of cost savings and commodity products, he said.
Motorola Mobility takes an agnostic view of cloud and does in fact look to move workloads among platforms when appropriate. It has a mix of AWS, GCP and Azure, along with its own OpenStack environment, and the company has put the onus on standardized tooling across platforms.
"If I can build an application at the simplest level of control, I should be able to port that to any cloud environment," said Richard Rushing, chief information security officer at Motorola Mobility. "This is kind of where we see cloud going."
Ultimately, a multicloud strategy comes down to IT shops' philosophical view, whether it's another form of a hosted environment, or a place to use APIs and put databases in order to take advantage of higher-level services, but can lead to lock-in, he added.
"I don't think there's a right way or a wrong way," Rushing said. "It depends on what you feel comfortable with."
Despite that agnostic view, Motorola hasn't completely shied away from services that tether it to a certain provider.
"Sometimes the benefit of the service is greater than [the concern] about what you want to be tied down to," Rushing said. "It's one of those things where you have to look at it and say, is this going to wrap me around something that could benefit me, but what else is it going to do?"
Experimentation and internal conversations about those tradeoffs can be healthy because it opens an organization to a different way of doing things, but it also forces developers to justify a move that could potentially restrict the company going forward, he added.
Cross-cloud not yet reality
A wide spectrum of companies has flooded the market to fill these gaps created by multicloud, despite some high-profile failures including Dell Cloud Manager. Smaller companies, such as RightScale and Datapipe, compete with legacy vendors, such as HPE, IBM and Cisco, and even AWS loyalists like 2nd Watch look to expand their capabilities to other providers. Other companies, such as NetApp and Informatica, focus on data management across environments.
Of course, the ultimate dream for many IT shops is true portability across clouds, or even workloads that span multiple clouds. It's why organizations abstract their workloads to avoid lock-in. It's also what gave OpenStack so much hype at its inception in 2010, and helped generate excitement about containers when Docker first emerged in 2013. Some observers see that potential coming to fruition in the next year or two, but for now those examples remain the exception to the rule.
The hardest path to span workloads across clouds is through the infrastructure back end, Colesante said. For example, if an AWS customer using DynamoDB, Kinesis or Lambda wants to move to Azure, there are equivalents in Microsoft's cloud. However, the software doesn't transparently allow users to know the key-value store equivalent between the two, which means someone has to rewrite the application for every environment it sits on.
Another obstacle is latency and performance, particularly the need for certain pieces of applications to be adjacent. Cisco has seen a growing interest in this, Compton said, with some banks putting their database in a colocation facility near a major public cloud to resolve the problem.
Alert Logic's data science teams are exploring what Google has to offer, but Colesante pumped the brakes on the cross-cloud utopia, noting that most companies are still in the earliest stages of cloud adoption.
"What you'd eventually like to get to is data science analytics on platform A, your infrastructure and processing and storage on platform B and something else on platform C," he said, "but that's a number of years before that becomes a reality."
Trevor Jones is a news writer with SearchCloudComputing and SearchAWS. Contact him at firstname.lastname@example.org.
See what drivers for multicloud model have changed
Learn all you need to know about multicloud
Reduce risk with multiple service providers