DMCA
Lazy Queueing and Direct Process Switch — Merit or Myths?
Cached
Download Links
| Citations: | 1 - 1 self |
BibTeX
@MISC{Elphinstone_lazyqueueing,
author = {Kevin Elphinstone},
title = {Lazy Queueing and Direct Process Switch — Merit or Myths?},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
The L4 microkernel, like many first and second generation microkernels, was designed to maximise best-effort performance. One component of its functionality critical to overall system performance is its interprocess communication primitive. L4 uses two techniques to minimise communication costs: direct process switching and lazy queue management. These techniques improve performance at the expense of real-time predictability of the scheduler. Now that L4 is being adopted in the embedded space, which features real-time requirements, we must determine if there is continued merit in using the optimisations. In this paper we quantitatively analyse the two optimisations using different kernel implementations and measure the performance improvements of the optimisations directly, and indirectly using the Re-aim benchmark suite. We find that the system-level performance improvements are marginal for this Unix-like workload. 1.
Keyphrases
lazy queueing direct process switch merit real-time predictability overall system performance direct process switching best-effort performance re-aim benchmark suite embedded space continued merit interprocess communication primitive unix-like workload communication cost second generation microkernels different kernel implementation l4 microkernel system-level performance improvement performance improvement lazy queue management real-time requirement
