Computing Reviews

An analysis of performance evolution of Linux’s core operations
Ren X., Rodrigues K., Chen L., Vega C., Stumm M., Yuan D.  SOSP 2019 (Proceedings of the 27th ACM Symposium on Operating Systems Principles, Huntsville, Ontario, Canada, Oct 27-30, 2019)554-569,2019.Type:Proceedings
Date Reviewed: 01/06/20

The GNU Linux operating system (OS) has become, for better or worse, the dominant OS for computer science (CS) research, for deploying large-scale Internet/cloud infrastructures and supercomputers, as well as for some small-scale smartphone and Internet of Things (IoT) devices. Linux evolution, since its introduction in 1991 by Linus Torvalds, has followed a Borg-like assimilation of technologies from other operating systems, as well as introducing its own myriad innovations contributed by ever-growing open-source and commercial technical and user communities. Due to such ubiquity, there is of necessity much ongoing research in Linux performance characterization and improvement. Here, Xiang Ren et al. characterize recent (in the past seven years) attempts to optimize the Linux kernel’s capabilities, efficiency, and performance, noting that some attempts at improving security and implementing virtualization have significantly slowed the OS and reduced scalability.

The authors introduce a benchmark suite to profile such historical OS updates and their impact on kernel operations, particularly poll functions and context switching; some updates such as context tracking almost doubled kernel entry and exit times. As a result of these findings, the authors recommend disabling or reconfiguring the new kernel enhancements. Doing so, however, might leave certain vulnerabilities unaddressed, such as Spectre and Meltdown. And most end users of the OS likely lack the necessary skills for such kernel hacking. So, while such changes to the OS internals can be valuable, attention to basic performance observation and tuning must also be addressed; an excellent source for such is Brendan Gregg’s BPF performance tools [1].

The paper admits to a focus only on Intel processors, although that certainly covers the vast majority of Linux implementations. Also mentioned is the difficulty of tracking and evaluating new kernel features that appear in Linux releases every few months. But attention to OS performance must be continuous, and the authors do provide helpful perspective and tools to assess the benefits and tradeoffs as Linux continues to evolve.


1)

Gregg, B. BPF performance tools. Addison-Wesley, Boston, MA, 2019.

Reviewer:  Harry J. Foxwell Review #: CR146826 (2004-0080)

Reproduction in whole or in part without permission is prohibited.   Copyright 2024 ComputingReviews.com™
Terms of Use
| Privacy Policy