PerfectScalePerfectScale

HANDS-ON OPTIMIZATION WORKSHOP 

Operating Kubernetes Multitenancy: Shared Cluster, Separate Headaches

September 29, 2026 at 11 AM ET

Save Your Seat!

Step 1 of 2

PerfectScale needs the contact information you provide to us to contact you about our products and services. You may unsubscribe from these communications at any time. For information on how to unsubscribe, as well as our privacy practices and commitment to protecting your privacy, check out our Privacy Policy.

Get one tenant's memory request wrong, and someone else's pod gets killed for it, on a bill that can't tell the two of them apart anyway.

Almost nobody runs one cluster per team anymore, and once you share a nodepool, requests and actual usage start drifting apart fast. That drift shows up as throttling, out of memory kills, and noisy neighbor incidents that platform teams end up debugging even though nothing in the platform is actually broken.

The same drift shows up in the cloud bill: one line for the node group, nothing for the tenant, built on whatever allocation model got cobbled together from those same bad requests.

And the problem doesn't stop at the cluster edge. The shared database, the cross-zone traffic, the inference endpoint every team's model hits, none of that has a tenant attached to it either.

This session dives straight into a live, multitenant cluster. You'll watch it happen: workloads right-sized autonomously, risk prioritized across clusters and namespaces, and multi-tenant costs attributed down to actual runtime consumption, no tags required. This is about running it and paying for it once you already have it.

What you'll walk away with:

  • See exactly how a single inflated request turns into someone else's 2 AM page, and why manual tuning doesn't cut it
  • Why it compounds at the node: padded requests wreck bin-packing, strand capacity no tenant will claim, and bill one team's safety margin to everyone
  • Where accountability stops at the cluster edge: the shared database, cross-zone traffic, the GPU pool and inference endpoint every team's AI workload hits, and what runtime-level attribution puts a real number on instead

Who is this session for:

Platform engineers, site reliability engineers, and FinOps practitioners responsible for shared Kubernetes environments.

Meet your speakers

Vikram Seshadri

Vikram Seshadri

Senior Evangelist for Enterprise Platforms , DoiT

Vikram Seshadri spent eight years in platform engineering in the fintech space, including a stint as Senior Vice President leading EKS platform engineering at Citi. Now a Senior Evangelist for Enterprise Platforms at DoiT, he helps engineering teams solve the hard operational problems of running Kubernetes at scale: reliability, cost, and everything in between.

Hili Paryenti

Hili Paryenti

Director of Product, Attribute by DoiT

Hili Paryenti is a Product leader with deep experience in technical product management, plus a background in backend development and data engineering, including as tech lead. Currently VP of Product Management at Attribute by DoiT, she has built B2B products across Unity, Luna Labs (acquired by ironSource), and Bidalgo. A true problem solver who loves people and technology, Hili also leads a women-in-tech community