Kubernetes Austin

Kubernetes Bin-Packing & Production Guardrails for LLM Apps

Capacity: 100
in-person
Event date
Sep 24, 26
06:00 PM - 08:00 PM CDT
Location
701 Brazos Street, Austin, Texas, US
About this event

🚀 Join us at Station Austin for Food, Drinks, and Kubernetes on September 24, 2026!

Station Austin (formerly Capital Factory) is generously hosting the Kubernetes Austin meetup!

Everyone is welcome whether you are new to Kubernetes or already experienced, come connect with the Austin cloud native community. We’ll share the latest updates, practical insights, and real world learnings from Kubernetes, CNCF projects, and modern platform engineering.

Featured Talks

🎤 Steve Fraser

The Bin-Packing Problem Nobody Talks About: What Actually Limits Pod Density in Kubernetes

🎤 Shiva Kalyan Reddy Giri & Nitin Garg

From AI Prototype to Kubernetes: Production Guardrails for LLM Applications

Expect good conversations, practical takeaways, food, drinks, and a strong community vibe.

📍 Venue

Station Austin 701 Brazos St, Austin, TX 78701

Arrival Instructions

Upon arrival at the building, please proceed to the 1st floor, Apollo room. We’ll be in the Apollo Room, 1st Floor (look for Capital Factory signs) after 5:45 pm.

Parking

We know parking in Downtown is tricky! So, you can park in the building garage for just $8.00 (validation parking tickets will be distributed)! Street parking will still be an option. More information on parking here: https://www.capitalfactory.com/parking/

🏢 Venue Sponsored by Station Austin

Thank you to Station Austin for sponsoring Kubernetes Austin! Station Austin is the center of gravity for entrepreneurs in Texas. They meet the best entrepreneurs in Texas and introduce them to their first investors, employees, mentors, and customers. To sign up for a Station Austin membership, click here.

🤖 Sponsored by Cast AI

Thank you to Cast AI for sponsoring Kubernetes Austin! Cast AI is the Kubernetes automation platform that keeps application performance and cloud costs on autopilot. It continuously analyzes how workloads actually behave and automatically tunes CPU, memory, and infrastructure provisioning — cutting cloud spend without hand-tuning your clusters. Learn more at cast.ai.

Looking forward to seeing you there!

Agenda
  1. 6:30 PM - 7:00 PM CDT

    The Bin-Packing Problem Nobody Talks About: What Actually Limits Pod Density in Kubernetes

    in-person

    Kubernetes bin-packing looks like a math problem: fit requests into node capacity, let the scheduler handle it. In practice, clusters rarely hit the density that math promises — and the real reasons have little to do with the scheduler itself.

    This talk walks through the actual limiting factors that cap pod density, using real (anonymized) cluster data:

    • Max pods per node: kubelet's --max-pods default and cloud IP allocation limits (e.g., ENI limits on AWS) often cap density before CPU/memory ever does.
    • Requests vs. actual usage: the gap between what pods request and what they really consume is the single biggest source of wasted capacity — and why teams over-request "just in case."
    • Topology spread vs. bin-packing: spread constraints and anti-affinity protect availability, but every spread rule is a direct tax on density. We'll unpack how maxSkew and zone imbalance actively fight packing efficiency.
    • Storage and networking limits: CSI volume-per-node limits, ReadWriteOnce contention, and IP/NIC exhaustion that quietly block scheduling nobody expects to be blocked.
    • Hidden overhead: DaemonSet tax per node, reserved resources (kube-reserved/system-reserved), and device plugin allocations.
  2. 7:15 PM - 7:45 PM CDT

    From AI Prototype to Kubernetes: Production Guardrails for LLM Applications

    in-person

    AI applications can move from idea to working prototype quickly, but deploying them reliably introduces challenges that model demonstrations often hide. Real-time APIs, WebSocket connections, external LLM services, vector retrieval, authentication, background processing, and unpredictable workloads all create new operational requirements.

    This session presents a cloud-native architecture for moving an API-driven generative AI application toward Kubernetes. Using a real-time AI assistant as the reference use case, the talk examines how to separate frontend, API, transcription, retrieval, orchestration, and persistence workloads into independently deployable components.

    Attendees will learn how Kubernetes primitives can support autoscaling, secrets management, health checks, resource controls, rolling deployments, workload isolation, and failure recovery. The session will also cover challenges such as scaling stateful WebSocket sessions, managing external model rate limits, tracing requests across AI services, controlling inference costs, and designing graceful degradation when a model or transcription provider becomes unavailable.

    The presentation is vendor-neutral and focuses on reusable Kubernetes and cloud-native patterns rather than a particular commercial platform.

Organizers