Table of Contents
RTEval
Rteval is a python program written to evaluate the performance of a realtime Linux kernel on a particular hardware platform. The program unpacks source code for two loads: hackbench and a Linux kernel compile, then loops running hackbench and a parallel build of the Linux kernel. While the loads are running, the cyclictest program is run to measure realtime performance under load. When the specified run duration is met, the loads are stopped and cyclictest outputs measured timer latency values in histogram format, which is analyzed by rteval. Rteval then writes an XML file to disk with information about the system (gotten through DMI tables), the raw histogram data collected during the run and the statistical analysis of the run. (Reference: https://git.kernel.org/pub/scm/utils/rteval/rteval.git/tree/README)
More information about rteval: https://git.kernel.org/pub/scm/utils/rteval/rteval.git/tree/doc/rteval.txt
Installation
rteval requires Python3 with the following dependencies:
- python3-distutils
- python3-dmidecode
- python3-lxml
- python3-ethtool
- python3-requests
- gcc 5.1 or higher, flex, bison, tar, xz (for kcompile load)
Most of the above Python3 dependencies should be available in package management systems for most operating systems.
The user must manually download and compile rt-tests. Compiling rt-tests additionally requires glibc and libnuma headers.
Following are quick instructions to install and run once all the dependencies have been installed:
$> git clone https://git.kernel.org/pub/scm/utils/rteval/rteval.git $> cd rteval # The following kernel will be used by the kcompile workload. A different kernel version can also # used here. But, corresponding changes will have to be made in KLOAD in Makefile and kernel_prefix # in ./rteval/kcompile.py before installation. $> wget -P loadsource https://www.kernel.org/pub/linux/kernel/v6.x/linux-6.6.1.tar.xz $> sudo make install $> sudo make runit
More information at: https://git.kernel.org/pub/scm/utils/rteval/rteval.git/tree/README
Results
Following is a sample result for the command:
Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate performance as measured by those tests. Any difference in system hardware or software design or configuration may affect actual performance.
=================================================================== [86/1801]
rteval (v1.38) report
-------------------------------------------------------------------
Test run: 2017-02-13 12:49:21
Run time: 0 days 0h 0m 11s
Tested node: vpatel-dev2
Model: -
BIOS version: (ver: , rev :, release date: )
CPU cores: 8 (online: 8)
NUMA Nodes: 1
Memory: 31.292 GB
Kernel: 4.4.2
Base OS: unknown
Architecture: x86_64
Clocksource: tsc
Available: tsc hpet acpi_pm
Load commands:
Load average: 4.07
Commands:
- kcompile: make -C /home/vpatel/anon_repos/rteval/run/rteval-build/linux-2.6.39 -j16
- hackbench: hackbench -P -g 16.0 -l 100 -s 100
Latency test
Command:cyclictest -qm -i 100 -d 0 -h 2000 -p 95 --smp
System: (8 cores) Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz
Statistics:
Samples: 869190
Mean: 2.12836433921us
Median: 2us
Mode: 2us
Range: 731us
Min: 0us
Max: 731us
Mean Absolute Dev: 0.937310248954us
Variance: 35.8090162273us
Std.dev: 5.98406352133us
CPU core 0 Priority: 95
Statistics:
Samples: 109626
Mean: 2.61308448726us
Median: 2us
Mode: 2us
Range: 645us
Min: 0us
Max: 645us
Mean Absolute Dev: 1.67699461734us
Variance: 90.2609495244us
Std.dev: 9.50057627328us
CPU core 1 Priority: 95
Statistics:
Samples: 109870
Mean: 1.86766178211us
Median: 1us
Mode: 1us
Range: 710us
Min: 1us
Max: 711us
Mean Absolute Dev: 0.921409903422us
Variance: 34.2588155195us
Std.dev: 5.85310306756us
CPU core 2 Priority: 95
Statistics:
Samples: 109463
Mean: 2.37517700045us
Median: 2us
Mode: 2us
Range: 621us
Min: 0us
Max: 621us
Mean Absolute Dev: 0.944711592396us
Variance: 31.7869418697us
Std.dev: 5.63799094268us
CPU core 3 Priority: 95
Statistics:
Samples: 107860
Mean: 2.1190246616us
Median: 2us
Mode: 2us
Range: 652us
Min: 0us
Max: 652us
Mean Absolute Dev: 0.876297487018us
Variance: 27.0425459294us
Std.dev: 5.20024479514us
CPU core 4 Priority: 95
Statistics:
Samples: 108567
Mean: 2.1447309035us
Median: 2us
Mode: 2us
Range: 684us
Min: 0us
Max: 684us
Mean Absolute Dev: 0.879960259054us
Variance: 29.0171586253us
Std.dev: 5.38675770991us
CPU core 5 Priority: 95
Statistics:
Samples: 108366
Mean: 1.66483029733us
Median: 1us
Mode: 1us
Range: 571us
Min: 0us
Max: 571us
Mean Absolute Dev: 0.755909668094us
Variance: 6.99546256844us
Std.dev: 2.6448936781us
CPU core 6 Priority: 95
Statistics:
Samples: 107590
Mean: 2.14418626266us
Median: 2us
Mode: 2us
Range: 731us
Min: 0us
Max: 731us
Mean Absolute Dev: 0.797404828348us
Variance: 37.2396828533us
Std.dev: 6.10243253575us
CPU core 7 Priority: 95
Statistics:
Samples: 107848
Mean: 2.0935761442us
Median: 2us
Mode: 2us
Range: 693us
Min: 0us
Max: 693us
Mean Absolute Dev: 0.844485575396us
Variance: 28.6649385662us
Std.dev: 5.35396475205us
===================================================================
rteval: exiting with exit code: 0
Most of the information is self explanatory. The statistics printed in the above example is just the output from cyclictest.
