Kernel Plugins: When A VM Is Too Much

Ivan Ganev, Greg Eisenhauer, Karsten Schwan
College of Computing
Georgia Institute of Technology
Atlanta, GA 30332
{ganev,eisen,schwan}@cc.gatech.edu

Abstract:

This paper presents kernel plugins, a framework for dynamic kernel specialization inspired by ideas borrowed from virtualization research. Plugins can be created and updated inexpensively on-the-fly and they can execute arbitrary user-supplied functions such that neither safety nor performance are compromised. Three key techniques are used to implement kernel plugins: (1) hardware fault isolation, (2) dynamic code generation, and (3) dynamic linking. Hardware fault isolation protects kernel-level services from plugin misbehavior, dynamic code generation enables rapid online creation of arbitrary plugins, and dynamic linking governs the kernel/plugin interface.

We discuss the design and implementation of the kernel plugin facility, as well as its advantages and shortcomings. Its use is demonstrated by a range of micro- and macro-benchmarks and a real-life application featuring plugins that dynamically transcode images served by a high-performance kernel web server. Benefits realized from plugins can be both qualitative (adapting services to clients' needs), and quantitative (improving performance through co-location of application plugin code with kernel services). Plugins are implemented in GNU/Linux on the Intel x86 platform. Reported performance results include plugin upcalls in 0.45-0.62 $\mu S$, dynamic code generation in 4 $mS$, and linking/unlinking in 3.1/1.6 $\mu S$ for an image grayscaling plugin - a dynamically code generated 66-line function written in a subset of C. All results are measured on an 866 MHz Pentium III.