Composing and Decomposing OS Abstractions

Abstract

Operating systems (OSes) provide a set of abstractions through which hardware resources are accessed. Abstractions that are closer to hardware offer the greatest opportunity for performance, whereas higher-level abstractions may sacrifice performance but are typically more portable and potentially more secure and robust. The abstractions chosen byOS designs impose a set of trade-offs that will not be well-suited for all applications. In this dissertation, we argue the following thesis: Supporting novel hardware such as non-volatile RAM (NVRAM) and new abstractions like fine-grained isolation while maintaining efficiency, usability, and security goals, requires simultaneous access to both high-level OS abstractions and compatible access to their low-level decompositions. We support this thesis by offering two new abstractions, PTx and light-weight-contexts (lwCs), as well as the null-Kernel, a new OS architecture. PTx is a new high-level abstraction for persistence built on top of NVRAM, a new form of persistent byte addressable memory, whereas lwCs are a new OS abstraction that enables fine-grained intra-process isolation, snapshots and reference monitoring. Due to the efficiency requirements of both PTx and lwCs, both abstractions required access to low-level decompositions of higher-level abstractions, while interoperabilityrequirements dictated that both low and high-level abstractions were exposed simultaneously. The null-Kernel is an OS architecture that enabled the simultaneous exposure of multiple abstractions for the same underlying hardware in a safe way, which, if adopted, would accelerate the development and deployment of abstractions such as PTx and lwCs.