Managed services providers (MSPs) who are looking to differentiate themselves in the shadow of hyperscalers, take heart: There’s always going to be infrastructure diversity. At the recently held infra/STRUCTURE presented by Structure Research in downtown Toronto, this diversity and opportunity is dubbed “the subscale alternative.”
While “sub” may have a negative connotation, panel moderator and Structure Research Managing Director Philbert Shih said that subscale can essentially be anything that’s not hyperscale. Without replacing colocation or other managed infrastructure categories, hyperscale has driven a lot of change. “Some of them are evolving; some of them are tailing off.”
Zachary Smith, the global head of edge infrastructure for Equinix, said it’s important to remember that most enterprises or even service providers look different than the oft-discussed hyperscalers because workloads are getting more specialized due to machine learning. “The applications that are hitting our lives are actually demanding silicon that is changing really rapidly and is really specialized for the workload,” he said.
This specialized hardware is going to be deployed in minimal quantities. In the meantime, workloads are getting more distributed, which needs a different kind of infrastructure than hyperscale, Smith added.
StackPath Chief Strategy Officer Ashok Ganesan prefers “decentralized horizontal scale” over subscale. Even as the company installs large numbers of compute nodes to provide unlimited capacity to applications in each location, a central cloud is always available as a fallback to meet capacity needs. He said it boils down to what kind of performance an application needs and how to simplify the experience — technologies such as Kubernetes and Continuous Integration / Continuous Delivery (CI/CD) pipelines are making it easier.
“If putting it in a central cloud is good enough for an application, there is no reason why they should distribute it, but if that’s not enough, you are better off actually breaking it down and putting the right components in the right places,” Ganesan said.
Ridge Co-Founder and Chief Evangelist Jonathan Seelig said the company’s platform is predicated on the idea that there is a huge amount of infrastructure not sitting inside of a hyperscale environment, regardless of what it’s called. The Ridge platform allows application owners to run managed services in many different places, including Kubernetes-managed, containers-managed, and object-storage services.
“In the vast majority of geographies all over the world, there is a fantastic service provider, generally more than one, which is not called Amazon, Microsoft, or Google, and who has the superlative space, power, and connectivity in that market,” Seelig said. The best place to run many services is one that’s been there long before Amazon Web Services, Google Cloud, and Microsoft Azure existed, he added.
Ganesan said that while it makes sense for certain aspects of a retail application to run in a central cloud, the best way to improve the user experience is to take advantage of resources wherever they are. Seelig echoes that approach but for the automotive sector, noting that automakers that have cars in 180 countries can’t rely on hyperscalers alone — decentralized capabilities in subscale environments are necessary.
Smith said standards are vital to taking advantage of subscale infrastructure wherever it’s needed so that it’s easy to tap into sustainable hardware anywhere, but standardization isn’t there yet, even though these subscale environments have been around for a while.
In the meantime, it’s possible that hyperscalers could address some of these more distributed workloads. Still, Smith says there’s a market for tier-two clouds managed by service providers offering subscale infrastructure that can be used by many people. Yet a lot of friction needs to be removed.
“We as an industry have got to make it easier so that those people can have business models, and we can have a few hundred highly distributed clouds versus only a handful of really big ones, which won’t service those niche use cases really well,” Smith said.