How to make hybrid cloud work

By:  Mary Branscombe

Last year, hybrid cloud was a priority for many enterprises, although with a certain amount of confusion about what it involved. This year, those ambitious plans are showing up as an approach to hybrid cloud that relies on platform as a service (PaaS) to take full advantage of cloud models.

The enterprise view of hybrid cloud is becoming clearer and more sophisticated than simply using cloud alongside existing on-premises IT, says Adam Warby CEO of managed services provider Avanade (a Microsoft/Accenture joint venture). “The way I’ve seen it evolve is that people have been getting more focused on hybrid IT than on hybrid cloud, and that’s relevant because most people are having to deal with managing legacy and existing applications while trying to reinvent what they’re doing as a business …. What we’re seeing now are different workloads that really do need the ability to work seamlessly between on-premises and in-public cloud.”


As an example, Warby points to the way you can run SAP workloads, including Business Suite and SAP HANA, on Azure and integrate services like Fieldglass, SuccessFactors and Concur with Office 365. SAP was a key workload for Avanade client Rio Tinto, “who are moving their entire IT as a service into the cloud,” Warby says. “It’s important we have that kind of flexibility. I think pubic cloud will be the dominant workload over time, but the ability to do close to the processor, high volume transactions close to the data is still going to be important.”

Distributed data for retail businesses is another area where Warby sees hybrid cloud becoming increasingly important. “This is the world of NoSQL and being able to do predictive analytics. Not ‘the sun was shining yesterday and I should have been selling ice cream’ but ‘the sun is shining today and I will sell ice cream, I’ll move them to the front of the stores.’ Most retailers are still doing trend analytics on data that’s 24 or 48 hours old.”

The new generation of on-premises applications that are designed for cloud integration, like the 2016 releases of SharePoint and SQL Server that connect to Office 365 and Azure services, are built for this kind of hybrid cloud solution, Warby says. “Cloud-native architecture that’s built to work on premise is actually the reality of how this has to work because ‘cloudifying’ on-premise architecture is a fairly difficult thing to do.”

Hybrid services and platforms

For CIOs, Warby says, the “hybrid IT” approach means becoming a service broker. “I don’t care if it’s in the public cloud or in our data centers, what are the services that run the company and are mission critical?” Starting with those services is the way to simplify hybrid cloud decisions. “When you start to work back from ‘What are the services?’ to ‘What’s available?’ [to] ‘What can I and can’t I do in public cloud versus on premise?’ then I think the data questions and the architecture questions become clearer.”


Enterprises are already used to this complexity in their existing business apps, which are becoming more distributed and tend to rely on a range of different services and platforms.

“When we ask people what platforms are you using, the answer is ‘yes’,” jokes Abner Germanow, senior director of strategic marketing at software analytics companyNew Relic. “If we ask ‘are you using Amazon, Docker, OpenStack, Cloud Foundry?’ the answer is ‘yes’. The list is a country mile long.”

It’s the same hybrid approach when it comes to business applications, Germanow says. “Not only are people using third-party services for infrastructure but they’re also using them in the front end, and they’re also using business partners to run specific services for them. The back-end service could be payment processing, it could be some computation, it could even be something coming from HR. Even for B2B applications, more and more they’re dependent on either other business units or on other third parties with service levels.”

That matches what Paul Veitch, senior director for technology services at Avanade, is seeing: The developers most interested in cross-platform .NET Core are matching it with Azure PaaS services.

Germanow says most cloud decisions as being about improving software. “When you move to cloud you need a short-term victory. Sometimes it’s performance gain, because your infrastructure was ten years old and this new one isn’t. But once you’re there, the expectation isn’t just that you’re saving money, it’s ‘where are my new apps?’.”
“”In order to do private cloud well, you have to be really good at software. You can’t say ‘we want to do private cloud and we’re also interested in getting started in devops’.”
— Abner Germanow

The level of sophistication in software that private cloud demands currently puts it out of reach for many businesses, Germanow says. “In order to do private cloud well, you have to be really good at software. You can’t say ‘we want to do private cloud and we’re also interested in getting started in devops’.” Germanow points to PayPal as one of the most successful users of private cloud, but adds that “they have 80 people who do nothing but work on their software layer.”

“Then there are companies who essentially have no other choice because they have hundreds of millions of dollars embedded in data center infrastructure, so for them to move faster they have to do some kind of private cloud,” says Germanow. “That’s where we see quite a bit of activity around Cloud Foundry and the various derivatives. But you have to be willing to make that fairly significant investment.”

Microsoft’s plan for Azure Stack

That software sophistication that is out of reach for many companies and the importance of services is what Microsoft is hoping to address with its Azure Stack hybrid cloud platform and explains why the first release of Azure Stack will be available as only an ‘integrated’ system with pre-built hardware from Dell, HPE and Lenovo.

Despite some disappointment among those who had hoped to use the ‘build your own’ approach, it turns out the majority of enterprises want Azure Stack to be not just a hybrid cloud option but the way that they get a version of Azure that includes all the Azure services — and the same velocity of updates and new servers as in the public version of Azure. And to make that happen, Azure Stack has to be as uniform as possible.

Microsoft was surprised by that appetite for the frequent updates that are common in cloud, when the on-premises approach to patches and updates has tended to be far more conservative. “We didn’t expect people to want to get the pace of innovation that we’re prepared to deliver,” Mark Jewett, senior director of product marketing for Microsoft’s cloud platform, tells
“”The thing we heard over and over is ‘How fast can I get full parity with Azure?'”
— Mark Jewett

While a small part of the market does want highly converged, integrated systems (spending on those is up 148 percent from a year ago, according to IDC), potential Azure Stack customers made it clear to Microsoft that they’re not looking for an integrated system for virtualization with a few cloud services; they’re looking for a way to get public cloud services with public cloud speed of innovation in places where they can’t or won’t connect to public cloud, Jewett says.

“The thing we heard over and over is ‘How fast can I get full parity with Azure? How fast can I get the 500 features you delivered last year in Azure? Because that innovation plan is part of what I’m investing in when I invest in Azure Stack.’ They want Azure services, but more importantly, they want this thing to be a living, breathing thing.”

“They told us ‘If you give me the five services I want, don’t stop there. If one of those services adds another innovative feature, I want to get it.’ The strength with which we heard that message was surprising to us and when you come back to the pace of updates, the reliability of updates is paramount. That circles back to being able to manage that and that requires a good affinity between the hardware and the software so that it’s a very smooth process,” Jewett says.

What Azure Stack customers are looking for is hybrid cloud rather than pure private cloud, according to Vijay Tewari, principal group program manager at Microsoft, who helped build the company’s hyperconverged Cloud Platform System. “Lots of customers are looking at a hybrid application model, where parts of the application are running on-premises and parts are consuming services from the cloud, and those could be SaaS applications or PaaS services. The whole app is a construct.”

“Our intention is to make sure we continue to provide innovative Azure services,” says Tewari. “To do that we have to make sure we have a robust hardware platform and we are not constantly burdened with trying to do matchmaking of compatible versions of firmware and drivers. That part needs to just work; that’s how we can get you all the cloud services. We don’t want our customers spending an inordinate amount of their resources — whether that’s time or money — just trying to keep the infrastructure running.”

Interest in Azure Stack isn’t about the cloud reluctance we’ve seen in the past; it’s about bringing public cloud to locations where a public cloud connection isn’t physically feasible. “We’re talking to oil companies who want PaaS-based apps but they will be sitting on an oil rig off the coast. That app necessarily can’t sit in any public cloud; it has to sit on the oil rig or on the ship,” Tewari says.

For other companies, latency, not geography, makes public cloud infeasible. “In a real-time operation, latency can have a real impact,” Jewett points out. “Any down time on the manufacturing floor is very expensive because everything stops. We’re hearing people start to talk about scenarios where they want to run manufacturing floor operations locally but still use public cloud for data processing and analysis more asynchronously.” That makes hybrid cloud a much bigger market, he points out.

Updates mean standardization

Microsoft has said it would be prescriptive about the hardware for Azure Stack; you were never going to be able to repurpose existing hardware, and servers with TPM 2 and RDMA NICs haven’t been available long enough for businesses to have spares lying around (although you will always be able to use your own hardware to build proof-of-concept systems).

Limiting the numbers of vendors limits the combinations of firmware and drivers that Microsoft has to test, not just for the initial deployment of Azure Stack on a particular hardware topology, but also for every single update — and there are going to be a lot of those to provide all those new and improved services. You just won’t ever have to care about them because Azure Stack will hide what would otherwise be a complex and lengthy process.

“With the Cloud Platform System, we’ve learned what it takes not just to deploy [this kind of system] but to update it,” says Jewett. Of the two, updating is more likely to cause issues. “What we saw with Azure Stack in the technical preview is what we saw in CPS. Frankly, it’s what we see on other vendors’ systems when you hear about where things are failing, and it’s why you see the growth of integrated systems as a category overall.”
“”The right way to get Azure into customer data centers is to make sure we have a fully fine-tuned hardware and software platform.”
— Vijay Tewari

“The thing people fixate on is just getting the initial deployment right, but doing the full operational life cycle is a much bigger proposition,” cautions Tewari. “That’s part of the reason why we took the decision and said that the right way to get Azure into customer data centers is to make sure we have a fully fine-tuned hardware and software platform.”

In the longer term, Azure Stack will come to more hardware than the initial three integrated systems (and Jewett notes that Microsoft chose their partners deliberately to keep pricing competitive). “Is it only going to be those three? The answer is no, that’s a starting point,” promises Jewett.

“There was a time when the Windows Server ecosystem was a fairly constrained set of systems and components that would work,” Jewett notes. “We think in terms of platforms and ecosystem, so you can well imagine that’s exactly how we’re thinking about approaching something like Azure Stack. We envision a world where we’re going to take the validation tools Vijay’s team is developing and increase what they support — but we start with those three.”

Patches and updates are also critical for security, points out Jeffrey Snover, lead architect for Microsoft’s enterprise cloud group and the Azure Stack. “Shocking as it is, it keeps coming up. People are getting hacked because they’re just not patching their systems and they’re not patching their systems because it’s a really hard problem. We are going to do that for them. We are going to patch their systems as a seamless update to the fabric layer in a way that preserves the operations. When we solve that, people are going to love that.”
Mary Branscombe

Mary Branscombe is a freelance journalist who has been covering technology for over two decades