Menu

I wrote an empirical research paper on Stateful Container Live Migration - would love to get your feedback and critique!

Reading 0:00
15s threshold

Hey everyone,

Stateless containers are trivial to move around - you just kill the pod and let the orchestrator spin it up somewhere else. But moving stateful workloads without downtime is still a headache.

I've been digging into this for the past few months and just put together a research paper benchmarking stateful container live migration. I mainly tested CRIU with pre-copy memory migration, looking at downtime, memory dirtying rates, and network switchover overhead under different loads.

The paper is up on Zenodo if anyone wants to check out the numbers:

https://zenodo.org/records/23779338?preview_file=Stateful_Container_Live_Migration_IEEE_NativeTeX.pdf

I’m really curious to hear from folks running prod clusters:

- Does anyone actually bother with container-level live migration, or does everyone just fall back to app-level replication and failover (e.g. database replicas)?

- If you've played with CRIU or similar tooling in production, what was the biggest pain point?…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.