Why bare-metal containers are scaring VMware [Q&A]
With a market capital of $73 billion and a stock price that has more than tripled in three years, VMware is one of the big hitters of the IT industry. The company has dominated the market since it introduced virtual machine server abstraction 19 years ago.
But some industry watchers cite the rise of containers as an existential risk to VMware. This new way of software deployment was popularized by developers via the Docker container format, and more recently is taking the data center by storm as enterprises take containers into production and standardize on Google’s Kubernetes orchestration platform.
We spoke with Tom Barton, CEO of container platform Diamanti, to learn why his company thinks bare metal x86 servers -- not virtual machines -- will prevail as the best infrastructure for containers, and why this is a major threat to VMware’s hold on the server market.
BN: What's driving enterprises towards containers in the first place?
TB: Enterprises are committing to faster, more flexible ways of building and delivering software. That means more frequent releases and constantly shipping new code. And that puts a strain on the old way of deploying software.
The old application stack is under siege with its three tier architecture; SANs, bloated middleware and application servers. Many of today’s most advanced distributed systems are built using frameworks that are deployed as containers, and built to run on clusters of bare metal servers. Modern Agile development emphasizes microservices architectures -- smaller components built to run in a more distributed fashion, built to handle failure and a more ephemeral nature, spin up and down as needed, and get real-time hardware provisioning requests out of the application workflow itself.
Containers and more specifically Kubernetes are also the path to hybrid workloads, the flexibility of deployment to on-premises or in the cloud or both. One of the main goals that people are looking for is application portability: freeing themselves up from vendor lock-in, being able to deploy in a number of different clouds and operating systems.
BN: At VMworld, Pat Gelsinger said that virtual machines are still the best place to run Kubernetes and containers. Do you agree?
TB: No, and I believe customers also know better.
First of all, VMware's acquisition of early stage startup Heptio for $550 million would suggest VMware is pretty eager to build some IP around containers, and figure out how to reduce the risk of eventual obsolescence of virtual machines.
VMs are fat. VMs require a lot of resources to sustain them. And importantly, inside each VM, you have to boot up an OS. So when you run containers on a VM, you're running a VM and an OS and a container, and that’s unnecessary overhead. Running each container on a VM also introduces two layers of orchestration--one to manage the virtualization environment, and then Kubernetes to manage the container environment. That means two layers of networking topology to have to figure out, and two layers of storage architecture to manage.
When you run on virtual machines, 40 percent of your infrastructure is supporting VMware and the associated resource obligations that VMware imposes.
A lot of cloud-native apps really are not meant to run on VMs, they're meant to run on bare metal. Most are I/O bound when it comes to performance. And everybody knows that with VMs, I/O performance -- and more importantly TPS (transactions per second) performance -- suffers.
BN: If containers will replace virtual machines, and if the enterprise datacenter standardizes on Kubernetes as seems to be happening -- what will the impact be on the vendor ecosystem?
TB: I believe there will be a time in the not too distant future where the majority of OEMs have Kubernetes-optimized x86 servers out-of-the-box, because that's the type of wholesale replacement you will see of VMs by containers.
We're on the cusp of a revolution the same way that virtual machines revolutionized the industry nearly two decades ago. If you were in the industry then, you remember that no one knew how to operate VMs. No one knew how to optimize storage in the network so VMs would work efficiently. It was all new. There were plenty who would run their VM hypervisor on whatever was lying around, so it was all far from optimal.
The first attempts at containerized infrastructure from the hyperconverged players have largely been about trying to squeeze containers into the virtualized approach. But the next generation infrastructure -- where startups like Diamanti are focused -- is about making it possible to run Kubernetes and containers on bare metal, the way the largest container production environments in the world (including Google) run their containers.