ZeroStack is a maker of fully automated infrastructure aimed, at least initially, at agile development environments. ZeroStack’s premise is to make infrastructure push-button, allowing IT teams to stand up their workloads in as easy a way as possible with as little setup as possible.
The turnkey private cloud system is aimed primarily at on premises use-cases, but can extend into the public cloud using connectors. AWS integration exists today, with plans for Azure and Google Cloud in the spring and summer, respectively.
What did they announce?
In the spirit of making infrastructure even easier to use, ZeroStack has added a learning capability to their policy engine. They bill this new feature as a “self-driving cloud.” And hey, why not? Autonomous driving is all the rage these days.
In a briefing, ZeroStack described the learning capability as “artificial intelligence” and “machine learning,” although I have developed an aversion to those terms. AI and ML are quickly becoming the new “software defined” to my ears. I am no longer willing to accept the description without qualification. Perhaps the terms apply here, and perhaps they don’t. Either way, the end result is certainly interesting.
What, exactly, is a self-driving cloud in ZeroStack parlance? It is a cloud that does the boring work for you. ZeroStack cites routine operations, management, and upgrades as tasks it can automate. The self-driving cloud is also a cloud that learns from you and your environment to improve resource utilization, predict future needs, optimize workload placement, and troubleshoot performance problems.
This sounds lovely, although some of these self-driving cloud features are on the roadmap. Workload placement optimization is planned for ZeroStack 3.0, while performance troubleshooting is meant for ZeroStack 4.0.
Today, ZeroStack considers itself to have arrived at hands-free operation — ZeroStack 2.0. Z-Brain, the learning engine that lives in the cloud, uses anonymized statistics, events, and health monitoring telemetry from customers’ ZeroStack clusters to learn how workloads are running. The engine then determines what can be improved. The workloads that are out of compliance with inferred global best practices are automatically “fixed.”
The data is analyzed a couple of different ways. For events with a short-term impact on the cluster, ZeroStack uses real-time processing with Apache Spark. Long-term decisions are driven by MapReduce calculations. The result is an automated, algorithm-driven insight into ZeroStack infrastructure. The decision making is precise, allowing for better utilization of the cluster hardware, as opposed to the typical over-provisioning done by most ops folks, who never want to hear from a workload again once it’s been released to the owners.
Hey, I’ve been there. Give them what they ask for, plus about 30%. You know, just in case they underestimated, since they always do…
ZeroStack shared the following specific examples of what their learning driven policy engine could do in the form of proactive remediation.
- Predict when a cluster will be out of capacity.
- Increase the size of storage volumes running out of space.
- Analyze storage performance and recommend disk type that would perform better.
- Automatically upgrade software.
- Observe underutilized virtual machines, resizing them more conservatively to free up hardware resources.
The view from the hot aisle.
Infrastructure automation is a hot-button item for organizations that need to make applications accessible in a hurry. You can call this “devops” or “agile” if you like. The point is that infrastructure — servers, storage, security, networking — is moving to the point of being easily consumable. Automatable table stakes. Deep skills in provisioning hardware will not be needed by organizations in the not-too-distant future, because software will be able to get the job done.
Don’t misunderstand. ZeroStack doesn’t intend to displace VMware, at least not right now. Think of ZeroStack with its self-driving cloud as a pod off to the side you turn over to your dev team to let them have at it. The processes and platforms built around VMware will be around for a while, because change is hard. But at some refresh cycle soon, an IT shop is going to start looking at folks like ZeroStack, and wonder why they aren’t automating some of the production tedium away.
After all, if it worked well enough through the DEV, QA, and UAT cycles, maybe it’s ready for production.