AI
The TAG Reboot: From Eight Groups to Five
Eleven TOC members cannot carry the technical vision for 240 CNCF projects alone. In June 2024, the TOC rebooted its technical advisory groups, cutting the count from eight down to five. Each TAG now gets two TOC liaisons. Working groups, which often dragged on with unclear scope, were replaced by initiatives (short-lived and time-bound) and sub projects (ongoing, requiring long-term stewardship). Anyone can join either structure, not only TAG members.
TAG Infrastructure: Six Technical Domains
TAG Infrastructure sits at the foundation layer, below workloads and developer experience. Its seven members, three chairs and four tech leads serving one- to two-year terms, cover storage, networking, DNS, compute, service mesh, infrastructure as code, load balancing, edge, and sovereignty. The focus areas are intentionally living. They were set at the reboot but can shift as ecosystem needs change. A booth at project pavilion P7A runs during morning hours at the conference.
What the Charter Actually Defines
The charter divides the domain into six building blocks: data, storage, networking, compute, infrastructure management, and edge/sovereignty. Compute now explicitly includes GPUs, TPUs, and FPGAs, not just CPUs. Infrastructure management targets zero-drift, auditable, self-healing systems from commit to runtime. The edge and sovereignty domain addresses intermittent connectivity, heterogeneous hardware, and regulatory boundaries that centralized data center patterns cannot handle one-to-one.
Three Active White-Paper Initiatives
The storage landscape v3 white paper defines attributes, including availability, scalability, performance, durability, and consistency, across topology types from centralized to hyperconverged. A separate AI/ML storage initiative examines training workloads (high throughput, high capacity) versus inference workloads (low latency, high IOPS via NVMe or GPU memory). The infrastructure life cycle initiative re-evaluates provisioning patterns for bare metal, edge, and private cloud, comparing reconciliation-based tools like Crossplane with event-driven tools like Terraform.
How to Contribute
Contributions range from attending meetings and posting on social media up to leading initiatives and becoming a tech lead. Project reviews are a concrete entry point: fresh reviewers catch things that experienced chairs already know to overlook. The TAG also wants help refining charter scope, identifying gaps, and surfacing industry trends. Contributors who get comfortable with the work can staff the project pavilion booth and eventually stand for TAG leadership roles.
Q&A
Does TAG Infrastructure produce reference architectures, or is that handled elsewhere? Reference architectures can come from a TAG (the former TAG App Delivery produced the platform maturity model this way), but the publicly marketed CNCF reference architecture page is driven by an end-user sub project and is separate. ▶ 27:31
Notable Quotes
two TOC liaison yay human load balancing Dylan Page · ▶ 03:07
the idea or the aim for tag infrastructure here is interoperability that any workload that uh should understand what a gold storage class means across vendors. Kashif Khan · ▶ 12:05
the goal is to have zero drift infrastructure which is auditable. We have uh self-healing and of course observable from comet to runtime. Kashif Khan · ▶ 16:26
a fresh perspective is always great uh when discussing these top high level abstract topics and white papers Dylan Page · ▶ 24:12
Key Takeaways
- Eight TAGs consolidated to five in June 2024, giving each TAG two TOC liaisons.
- TAG Infrastructure covers six domains including compute hardware like GPUs, TPUs, and FPGAs.
- Three white papers in progress address storage attributes, AI/ML storage, and infrastructure life cycle across bare metal and edge.
About the Speaker(s)
[Kashif Khan] is a co-chair of CNCF TAG Infrastructure and has maintained the CNCF project Metal3.io for six years. He works as an open source architect at Ericsson Software Technology in Finland and holds a PhD in Computer Science.
[Dylan Page] manages the Core Infrastructure team at Lambda.ai and co-chairs both CNCF TAG Infrastructure and Kubernetes SIG Infra. He also maintains Atlantis, a CNCF sandbox project, with a current focus on bootstrapping and scaling bare-metal Kubernetes for AI/ML datacenters.