What Is Disaggregated Kubernetes? Shrinking Blast Radius

What Is Disaggregated Kubernetes? Shrinking Blast Radius

WTF is Disaggregation?

Disaggregation means taking something that has traditionally been bundled together and splitting it into smaller, more independent pieces. Imagine a hotel where one master key opens every room, the kitchen, the electrical room, the safe, and the front desk. That's super convenient, but if someone steals that key, they can potentially get everywhere. That’s the problem we’re looking to solve with disaggregated Kubernetes.

A traditional virtualization architecture can have a relatively large list of privileged workloads responsible for a lot of things. Those workloads need to perform a whole bunch of different jobs from networking through to storage and device management. Edera's approach is to break those jobs apart into separate services/trust domains, so that the workload sits in a zone isolation boundary that is separate from the storage, network, and device infrastructure services.

The important part isn't merely that these are separate processes. It's that Edera can give each one a restricted view of the system and restricted authority. So, for example, the storage service might be allowed to access particular pieces of guest memory that have explicitly been shared with it. It doesn't need blanket access to everything the workload is doing. That's where disaggregation becomes a security concept rather than just a software architecture concept.

Why Edera Cares About Disaggregation

Suppose there's a vulnerability in the code handling network packets. In a highly-integrated architecture, the vulnerable component might have enough privilege to reach a lot of infrastructure or other workloads. 

With disaggregation, the intended model is more like a network bug is found, and it’s able to compromise network service. However, the network service is essentially contained. The bug still exists, but it cannot escape to other workloads or infrastructure. It doesn't mean the network service becomes magically harmless. If it has excessive authority, an attacker can still do damage. That's why we make the important distinction between isolation and authority.

In the case of agentic AI workloads, you need both. You need disaggregation to ensure a specific component is separated from the others. But you also need the capabilities (least privilege controls) to dictate that a specific component is only allowed to do a few very specific things, like access to github or to a tool like curl. Put together, if something gets compromised, there are fewer things it can reach and fewer things it is allowed to do.

So, WTF Is Disaggregated Kubernetes?

In this case, we're essentially taking the same idea up the stack. We don’t look at it from the perspective that Kubernetes has a privileged node agent that can do lots of things, instead we ask ourselves why our components need to have all of those permissions/capabilities. 

So, perhaps, networking is one capability, storage another, image handling another, streaming another, attestation another, and so on, and so on. Instead of giving one giant component the keys to the whole hotel, we're essentially trying to give each service the particular key it needs for the particular room it needs to enter at the Kubernetes-layer. Edera applies disaggregation to infrastructure by separating things like storage, networking, device emulation, and node-agent functions into independently isolated services with narrowly-scoped access, reducing the blast radius when one of them is compromised.

FAQ

What is disaggregated Kubernetes?

Disaggregated Kubernetes splits infrastructure functions such as storage, networking, device emulation, and node-agent work into separately isolated services, each with narrowly scoped access. Instead of one privileged component holding broad permissions, every service gets only the authority it needs, so a compromise in one place has a far smaller reach.

How does disaggregation reduce the blast radius of an attack?

Disaggregation gives each infrastructure service a restricted view of the system and restricted authority. If a bug is found in, say, the network service, the fault is contained to that service. It cannot automatically escape to other workloads or infrastructure, which shrinks the blast radius even though the vulnerability still exists.

What is the difference between isolation and authority?

Isolation separates a component from others so a compromise stays contained. Authority governs what that component is allowed to do. A service can be isolated yet still cause damage if it holds excessive privilege, so Edera pairs disaggregation with least-privilege controls to limit both reach and permitted actions.

Why do agentic AI workloads need both isolation and least privilege?

An agentic workload needs disaggregation so each component is separated from the rest, and least-privilege controls so a component can only do a few specific things, like reaching GitHub or curl. Together, if something is compromised, there are fewer things it can reach and fewer things it is allowed to do.

Cute cartoon axolotl with a light blue segmented body, big eyes, and dark gray external gills.

You know you wanna

Let’s solve this together