Conversation

@elmiko @djsundog anything that expects very low latency between network traffic and disk activity (i.e. serving database requests in <5ms, preferably <1ms) you have to worry about pinning interrupts and processes to specific cores. Beyond this you have to do careful planning of how bw is allocated between local storage (again, latency, not bandwidth, so SANs are unacceptable) and processes runnign on machines to avoid the "noisy neighbour" problems.

Way beyond what k8s/swarm can do today.