Previous OS abstractions and structures are mainly proposed for the average performance. The shift toward server side computing calls for new OS structures for the worst-case performance. This paper presents the isolate first, then share OS architecture. We decompose the OS into the supervisor and several subOSes running in parallel: a subOS directly manages physical resources without intervention from the supervisor (isolate resources first), while the supervisor can create, destroy, resize a subOS on-the-fly (th…