To establish the results, we follow the research methodology
developed in our earlier work, extending the software infrastructure
as necessary. Our development environment consists of a
parallelized z-kemel [8, 161 running on a SGI Challenge multiprocessor.
We conduct a set of multiprocessor experiments with
the I-kernel’s UDP/IP/FDDI protocol stack, measuring packet
processing times under specific conditions of cache state. The
measurements are used to parameterize the analytic component of
a simulation model of multiprocessor networking, under various
affinity-based scheduling policies. The analytic model captures
the displacement of the cached protocol state by the background
workload, and the resulting impact on packet processing time.
We begin with a review of the problem formulation in Section
2, and relevant aspects of the research methodology in Section
3. New performance results are presented in Sections 4-8, and
summarized in Section 9. Due to space limitations, we omit
a review of the related work in affinity-based scheduling; see
[20,22] for discussion.