Plurio: Scaling Microkernel-based Operating System via Service Localization and Memory Pooling

Abstract: Microkernel-based operating systems offer strong isolation but often scale poorly on multicore platforms due to expensive IPC. We present Plurio, a microkernel-based operating system architecture that revisits service organization and memory management around two principles: localizing service execution and pooling memory resources. Plurio dynamically combines tightly coupled services to turn cross-service invocations into local function calls, and employs a global memory pool that exposes physical memory as capabilities for mapping in a shared virtual address space. Affinity-aware worker arrangement and proxy-managed request caches further shorten communication path and reduce data copying. We implement Plurio on the Genode framework and evaluate it with various workloads, with experimental results showing that Plurio improves performance by up to 5.1× over existing microkernel systems on multi-core platforms.