Headed to AWS re:invent 2018? Here’s everything you need to get up to speed on the latest in Enterprise Cloud.
Last year, I published a recap of enterprise infrastructure trends ahead of Amazon re:Invent. This post provides an overview of the latest happenings in cloud infrastructure and major themes we’re seeing through our network of infrastructure tech leaders. Some of these thoughts are captured in more detail in our 2018 Enterprise Almanac,
Through the many roundtables, lunches, and events that we’ve hosted through corporate engagement at Work-Bench, we’ve observed an acceleration in workloads moving to the cloud and a total shift in attitude from the business, pushing teams away from the undifferentiated heavy lifting of building IT scaffolding and towards usage of higher order abstractions with growing open source platforms and cloud providers. The flexibility of Kubernetes has made it the universal standard for building and running modern applications. We’ve been following this trend for a while, having previously invested in CoreOS, the first company to introduce an enterprise Kubernetes product, which led to their successful sale to Red Hat (and now IBM). We believe that the entire lifecycle of the application will be managed by Kubernetes and are excited by the components that will be built on top of it.
Serverless Kubernetes is imminent
Kubernetes is now available from all of the cloud providers, as well as the likes of VMware, IBM, Cloud Foundry, and even Docker. It was built to provide clean APIs that abstracted away management details of machines, but at the end of the day the machines are still very much there. The concept of “serverless” allows people to forget about servers entirely, and the integration of Kubernetes orchestration with serverless container infrastructure is critical to shift workflows from running containers to running code, freeing up management responsibilities. As such, every cloud is undertaking efforts to bridge serverless containers with the Kubernetes API. Most notable here is Virtual Kubelet, and my suspicion is VMware may provide something soon given its latest acquisition of Heptio. I expect Amazon will introduce a number of new improvements to their Kubernetes service, EKS, to make the experience smoother and more consistent with other products in its arsenal. Most anticipated here could be an announcement for FarGate support on EKS, auto-scaling containers that are run in Kubernetes using a pay per use model. Whenever there’s a useful abstraction, you can count on the cloud vendors to package it up and run it as a service.
Key takeaway: I predict the pay per call serverless business model will just be a feature of the cloud and all clouds will have an offering a serverless Kubernetes offering. The ultimate value in the end will lie in performance, cost and interoperability.
Enterprise cloud is cloud native
In the last decade, we’ve seen startups leverage the cloud to develop and deliver new business models. The next ten years is about the enterprise cloud and will unlock a whole set of new values. Similar to the app store on mobile devices, a uniform set of standards can enable companies to better leverage their IP and share services across critical applications. These new businesses will be driven by digital ecosystems and the industries that have most potential are those that are rich with data — like financial services, health sciences, and manufacturing. The cloud is the optimal platform where components can be glued together and connected to services bringing in different forms of ML and AI. Some of the first movers to the cloud, however, are having a bit of a hangover. Many mistakenly focused just on cost savings and some of those architecture and vendor-related decisions are coming back to bite them. However, we’re seeing more executives from the Fortune 1000 now realizing that the real potential of cloud is actually the new business models and delivery mechanisms that are enabled. Folks with this mindset will create momentum for their digital ecosystems, and this could become a watershed moment for some winner takes all opportunities in different industries. The stakes are high and the race is on for new paths for revenue and monetization. In financial services, multi-party machine learning models trained by customer data is an ever present reality. I think that one really good example can become a catalyst here and I’m looking forward to seeing that within the Fortune 1000 by 2020.
Key takeaway: The cloud is less about cost savings and managing datacenter resources, and more about new business models and value delivery for end customers. We need the appropriate guardrails to do this securely, and these components are coming out of the cloud native ecosystem and new startups like ConcourseHub.
Data & AI
The clouds have all released products and services for the development of ML and even deep learning models. Azure ML and SageMaker are very similar in core functionality, but the former delivers more flexibility and covers more bases for both beginners and experts. I expect Amazon to make it easier to bring your own model, which today is a bit of a tedious process, as well as announce improvements like pre-built industry specific algorithms. Microsoft, Google, and IBM have also been building on AutoML, and Amazon will likely join the party. With that said, fundamental challenges still exist. While Docker and Kubernetes made it easy to build components, assemble, and upgrade them easily in production, the same set of advances need to happen for data pipelines, DAGs and algorithm management. I’ve written about some of the challenges in working with data today along with a landscape of startups solving for them, investing in best of breed tools that automate key parts of the AI process will help enterprises recruit and upskill talent.
Key takeaway: The biggest challenges here are engineering related on both sides of building a data science model. Getting, understanding, and transforming raw data and productionizing data science models across clouds in a serverless fashion are two top challenges that emerging companies, Datalogue and Algorithmia, are solving for and it requires a developer first approach to be popular with modern data workers.
A re-think of teams and processes is underway
Three quarters of CIOs around the world believe that IT has become so complex that it threatens their ability to effectively manage digital performance. IT has traditionally operated in silos, and the emerging cloud native landscape requires continuous coordination between teams, which has significant impact on organizational structures and operating models. How do new and emerging roles like FaaS, services & events engineers fit into the existing talent pool? How do you manage dependencies in a serverless environment? How are applications being tracked and managed over time? A lot of questions exist.
A key role that’s emerging in the development of modern cloud stacks is the Site Reliability Engineer (SRE). Although this role has existed for years, a hybrid sysadmin and developer, SRE best practices from web scale companies are starting to make their way to the enterprise because complexity becomes exacerbated when utilizing microservices and distributed environments. Monitoring and distributed tracing play an important part as root causes for a failure may be due to a recent change or a downstream dependency. A culture of continuous learning, investigative work and positive reinforcement is required. Without it, operations teams risk burnout given high stress finger pointing that tends to happen during a major outage.
Key takeaway: Creating a culture of transparency and using tools like Slack help create lightweight internal records of actions taken, so teams can resolve incidents quicker, loop in the right level of expertise when necessary, and use accurate context to quickly ship a post mortem. Analytics captured on incidents feed into broader reliability metrics for each application, like SLIs and SLOs, and how Error Budgets are affected. Process, documentation, and automation will be crucial going forward.
Identity crisis
Most of the world’s business applications were designed based on assumptions of trust, a reason why the most critical softwares are running in private networks. Cloud mindset is less about building and maintaining infrastructure but instead about building and maintaining access and control. It’s not about capacity or serverless, per se, but the real value to the business is being able to rethink digital interactions through standards of secured communications. This shift in mindset needs to happen.
We have a situation where talented people are constrained to experiment and learn, primarily because of security concerns. I’ve actually seen infrastructure executives who are passionate about their craft build and run their own “lab” at home to learn what’s possible with distributed cloud systems. There are open source technologies and cloud services that are backed up with rich, interoperable APIs, but talented engineers are hamstrung by the ability to experiment with them. In IT, there are segregated degrees of control, ownership and policy. How do you support the ability to empower people that want to get their hands on the latest tech? Aside from rethinking teams and processes, this requires a disintermediation of the historical IT control model. We are seeing teams opt for built in security versus bolt on, meaning engineering teams are taking on more responsibility when it comes to security. First order of security is that you can’t secure what you don’t know and therefore the biggest battleground today is identity and the zero trust model. The ephemeral nature of containers spun up and down according to demands and connecting to other services makes it hard for larger organizations to authenticate each interaction.
Key takeaway: The open source project, Spiffe, used to establish workload identity is becoming the defacto standard to embed identity across environments. The team behind this project, a startup named Scytale, is working with customers large and small to help them deliver customer value faster in dynamic IT infrastructure.
I’m excited by all that re:Invent and the future of cloud infrastructure holds for large companies where employee populations are getting excited about delivering new digital experiences similar to startups that have led business model innovations in the cloud for the past decade. This post was written with some haste, so would love to know if you agree or disagree with some of these thoughts. If you’re an entrepreneur or an executive passionate about some of these challenges, we’d love to chat.
Note: ConcourseHub, Datalogue, Algorithmia and Scytale are Work-Benchportfolio companies