Why up-sizing nodes usually doesn't fix Kubernetes P99 spikes

Market insights

Stack

Apply

Upload your CV — we will connect you with the employer directly through our pool.

Description

Why up-sizing nodes usually doesn't fix Kubernetes P99 spikes Lately, I’ve been looking at large clusters where the default answer to P99 spikes is vertical scaling. Teams throw more cores at the problem to give apps room to breathe, but it often fails to solve the root cause. We're testing a layer that allows the kernel to prioritize execution based on the specific runtime needs of each workload. Instead of treating a critical database and a background scanner the same, we give the kernel the context it needs to prioritize execution in real-time. In our lab tests, P99 latency for Redis and Nginx dropped by about 85 percent and database throughput increased by roughly 60 percent. This happens beneath the app layer, so there are no sidecars or code changes. I’m curious if this resonates with your experience. Do you up-size nodes just to stabilize graphs even when utilization is low? Would a read-only report showing exactly where your node is fighting your hardware be useful for your team? We are looking for one or two real-world environments to validate our data. We have a non-intrusive Observe Mode that just monitors signals and generates a report without changing any scheduling. If the data shows clear potential for improvement, the logic can move into an active mode to fix those bottlenecks automatically in runtime. Feel free to ping me if you want to chat or see the technical benchmarks. I’m keeping this anonymous for now due to current contracts, but would love to hear more about real use cases and pains! [link] [handle]

Employer contacts (email/phone/telegram) are hidden from the public preview — send your CV, and we will connect you directly.