Skip to content
InfoQ·

🎬Netflix Runs Single Tenant Titus Containers on Dedicated Right-Sized EC2 Instances

Your noisy neighbors just got their own instance

TL;DR

Netflix added single-tenant Titus, which runs one container alone on its own right-sized EC2 instance. Stateful, high-throughput streaming workloads are first in line. If you run latency-sensitive batch or streaming jobs on shared Kubernetes, this is the pattern to steal.

Netflix just gave its noisiest containers a room of their own. The new single-tenant mode of Titus runs one task per EC2 instance, sized to fit that task's CPU, memory, and GPU request. Shared kernels can't isolate a container that floods the host with small packets, so Netflix stopped trying. If you run a shared Kubernetes platform, pay attention. Application teams keep the same API and deployment path, while the Compute/Platform team still owns and patches the host. That split is the whole point. Teams get isolation without taking on fleet sizing or OS patching. The pattern is portable too: dedicated node pools with taints, pod anti-affinity, and repacking turned off. The numbers matter. Titus launches millions of containers a week across tens of thousands of EC2 instances. Fresh instances need tens of seconds to boot before the image pull even starts, so launch latency is the real cost. Netflix keeps a small warm pool for the most latency-sensitive starts and limits the sizes it offers. Netflix recommends this mode for large, long-running, disruption-sensitive workloads, not small or bursty services.

Netflix Runs Single Tenant Titus Containers on Dedicated Right-Sized EC2 Instances — InfoQ

Key Points

1

Titus launches millions of containers a week across tens of thousands of EC2 instances, which is why the single-tenant mode had to stay operationally invisible to application teams.

Keep reading — it's free

Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.

Also get
or

Already a member? Sign in

NetflixTitusEC2KubernetesPerformance Isolation

Comments

Subscribe to join the conversation...

Be the first to comment