As cloud evolves, why aren't more companies migrating?
How do you optimize a migration plan to ensure optimal cost savings in the cloud? And why aren't more companies migrating?
To address these questions, I’ve invited Aaron Rallo, CEO of TSO Logic, to share his perspective. TSO Logic offers an analytics platform that fast-tracks cloud planning and transformation. The platform has been used to model cloud migrations for more than 250,000 instances across dozens of enterprises. Aaron and I have known each other for a while, and I thought he’d be perfect to share his perspective on cloud migration. Here's our interview:
Welcome Aaron! Let’s dive in. Given the huge interest in cloud migration and the number of customers who have identified it as a top priority, why aren’t more companies migrating? What’s holding companies back?
According to our metrics, there are currently 225 million servers running in data centers today, so we’re obviously still at the forefront of the cloud migration trend. We see more and more workloads moving to cloud, but the reality is, it’s still just a small percentage of what’s out there.
So, what’s holding enterprises back? The biggest barrier is time. It takes time to assess your portfolio, to figure out the right cloud fit, to get to the low-hanging fruit and create a culture around cloud. For enterprises that have been doing things the same way for many years, cloud requires a new mindset and new tooling. It just doesn’t happen overnight. But it will happen—the rewards are just too great. We’re seeing savings in excess of 30% when moving from on-premises to cloud.
That makes sense. I still see lots of people challenging the idea that cloud is always cheaper. What factors do you see driving migration one way or the other?
I can see how some would think that. The fact is, getting accurate numbers for what you can expect to pay when migrating is still almost comically complex.
There’s not just one option called “cloud” with a price tag attached; cloud providers have massive catalogs filled with nuances to cover every possible scenario. We’re talking about hundreds of thousands of different compute and software combinations. Trying to manually identify the best fit for each workload is enormously difficult. And if you choose wrong, cloud will look a lot more expensive than it should be.
That’s really the core problem we set out to solve: how can we address this algorithmically, so you can get to concrete answers without spending hours and days wrestling with spreadsheets? One of the ways our platform finds the “best fit” is to analyze on-premises workload patterns of compute, software, and storage, and then analyze what that would cost to run in the cloud. It does this using both “like for like” provisioning—deploying instances in the cloud exactly as you have them deployed on-premises—as well as “optimized,” buying only the cloud resources you actually need based on your historical utilization.
What’s become clear is that most organizations don’t realize just how overprovisioned they are. According to research we published last year, more than 80% of on-premises resources are currently overprovisioned. So, when you look at “like for like” provisioning, cloud often costs the same or up to 10% more than current on-premises costs.
But when you optimize for what you actually need, all of a sudden cloud generates savings of 26% or more, on average. Now, start factoring in things like software license mobility and re-platforming to take advantage of cloud-native services—things that are even harder to account for when you’re trying to get to these numbers manually—and the savings grow even more.
Even beyond that, we’ve seen lots of cases where customers are also vastly underestimating the potential benefits of the cloud. Some of these benefits are harder to see in a direct infrastructure comparison – labor efficiencies, application refactoring, time-to-market improvements, Capex to Opex shifts, etc.
That’s correct. And another major factor that can be really opaque in these analyses is software licensing. For example, are you going to bring your existing Microsoft licenses to the cloud or buy new licenses from the cloud provider? Both Azure and AWS can accommodate license mobility, and avoiding the cost of new licenses can make a huge difference—80% or more in monthly costs, depending on the application. But even this is not a straightforward question. Not every instance supports license movement, and there is often more than one way to do it. For some workloads, you actually end up saving more by buying new licenses.
You also have to factor in the potential cloud savings for software with licenses based on core count. If you’re running that application on a four-year-old server that needs more cores to do the job, you’re actually paying a licensing premium to use older hardware. When you move to a newer cloud instance with more powerful compute, you might be able to run that workload with half the cores, which will definitely reduce your overall licensing costs.
Shifting gears a bit, how do you think about multi-cloud usage in a cloud migration model? We’ve seen that most customers don’t start with a one-size fits all approach– some workloads make sense on AWS, some may be on Azure or GCP. What have you seen?
This is a great question because it really speaks to the way organizations are evolving their use of cloud. There are two perspectives. First, there’s infrastructure-as-a-service (IaaS), the original cloud model and the lens through which many enterprises still view cloud. For organizations chiefly focused on IaaS, they may evaluate multi-cloud, but they typically choose a single cloud when it comes time to migrate.
This is understandable—if you’re adopting multi-cloud for infrastructure, you’re looking at a steeper learning curve and more significant support requirements to use multiple platforms. That’s especially true when you have centralized IT. And it’s often not worth the hassle.
Increasingly though, we’re seeing enterprises looking at how cloud can help them offload some of their day-to-day application maintenance responsibilities, in addition to infrastructure maintenance. So, they’re looking more closely at platform-as-a-service (PaaS) options. This could be as basic as moving infrastructure to AWS while moving Exchange to Office 365. Or, it can get into more fundamental choices in how enterprises write and manage their applications, so they can take advantage of things like cloud database-as-a-service platforms like RDS instead of running SQL Server on-premises.
Once enterprises start down the PaaS path, they tend to take on more of a “best tool for the job” mentality. At that point, multi-cloud strategies start to make a lot more sense.
Part of what makes cloud so interesting is how quickly providers like AWS are building new services and bringing them to market. You did some research recently after AWS announced their new C5 instance types. Can you talk about that?
One of the great things about having a data model with billions of data points is that you can have some fun with analytics. As a little side project, our team took the cost of running a workload today and compared it to what it would have cost to run that same workload in the cloud five years ago. Over that span, we found that the cost of cloud has decreased significantly. From just 2014 to 2017, for example, the cost of running the same set of workloads on AWS has come down by almost 75%.
Now, was that just a blip, or can we expect it to continue? To see if those gains are still coming, we took a deeper look specifically at the AWS catalog, comparing the cost of cloud before and after the introduction of their newest C5 compute resources. We found the C5 would be a better match for almost 20% of instances in our data set, and that migrating to C5 instances would generate annual savings of 18.5%. In hard numbers, if you’re currently spending $2 million per year, migrating to the latest, greatest AWS instances can save you $360,000 annually. So that downward cost curve is still going strong.
So once a customer understands that initial business case, what’s next? We know that customers don’t migrate all at once. Is there a process you’ve seen that they follow? How do they know if the cloud migration was the right decision?
One of the more common trends we’re seeing is that enterprises target migrations around infrastructure consolidations. For example, they have a datacenter that’s coming up on the end of a contract. Or maybe they have aging hardware and need to go through a refresh cycle. When you tie your migration to these kinds of events, you can collapse the costs more quickly.
But still, how can you be sure you’re doing the right thing? Here, KPIs are essential. We are seeing enterprises benefit from migration from the perspective of both cost and scale. Having accurate benchmark data to compare against, and concrete numbers for workload profiles and utilization as your cloud portfolio changes, becomes very important.
Since the cost of cloud is not static, we’re also seeing financial stakeholders getting more involved in cloud decision-making. They’re measuring cloud costs and starting to ask the hard questions of IT to ensure spend is on track or under budget. I think we will see a continued trend towards transparency into compute usage and cost, along with more guardrails imposed by financial stakeholders. All of this underscores the need for cloud analysis as an ongoing activity—not just something enterprises do once every few years.
The value management story is really interesting to me. We’re going to spend some time in a future blog talking about some of the metrics organizations can use to track the efficacy of their migrations – but obviously, they have to get started somewhere. How do you see customers taking that first leap?
One of the key metrics we’re seeing is around application trending reports. These essentially evaluate the cost of running an application before, after, and during migration, and identify things that could be done to further decrease costs. The trick to all of this is that enterprises will likely have a “migration bubble” as they transition to cloud. Throughout a migration project, they’ll still have the cost of maintaining an existing data center, and the bulk of those costs can’t be alleviated until the very last server comes out. Once again, accurate financial tracking and KPIs are essential to ensure you are on track and have full visibility into all your costs.
Let me see if I can sum up. We haven’t yet approached the tipping point for cloud, but enterprises are moving there more quickly as costs keep declining and cloud capabilities continue to evolve. And the further we get along this path, the more we should expect cloud migration decision-making to be governed by hard numbers and KPIs.It’s definitely a fascinating transition we’re in the midst of. We’ll check in again soon to see how things are shaking out.
For more on cloud migration strategies from Josh Heller, read An economic framework for cloud migration.
Howard Rubin on IT Benchmarking and TBM
Howard Rubin President and CEO, Rubin Worldwide