A Rust wrapper for Intel's Running Average Power Limit (RAPL) present in Intel CPUs since the Sandy Bridge (2nd) generation.
This wrapper functions mostly as a benchmarking tool, however, it also allows for performing live measurements with terminal output. By default the entire system consumption is measured, however, it is also possible to estimate pure software consumption by offsetting the benchmark consumption based on idle system consumption.
Due to inaccessibility on Windows and MacOS, raplrs only runs on Linux systems.
As such the following steps are only suited for Linux.
In addition to the Rust dependencies listed in Cargo.toml, rapl.rs has an OS-level dependency used for measuring CPU temperatures: sensors.
This dependency is contained in the lm-sensors (Ubuntu) and lm_sensors (Arch) package.
# set exec permissions on scripts
$ chmod -R +x benchmark/
# build
$ cargo build --release
# run
$ sudo ./target/release/raplrs
# for ease of access you can symlink raplrs to /usr/bin - preferably with absolute paths
$ sudo ln -s /path/to/target/release/raplrs /usr/bin/raplrs
# and then run with
$ sudo raplrs
If you are running a shell script that calls some command you would usually run from your command line, you need to make sure that said command is present in your /usr/bin dir.
For example, crispy-doom on Ubuntu by default installs to /usr/games, which is not recognized when running raplrs as root.
As raplrs mostly needs to be run with root access, you need to ensure that any configuration for the benchmark is present where it needs to be.
That is, under / rather than your user directory.
Scripts are located in ./benchmarks/. We discern between three types:
interactive: an application that expects input, e.g., a GUI.stdoutandstderrshould be piped to/dev/null, see./benchmark/interactive/cura.shfor an example.macro: larger benchmarks that perform several tasks../benchmark/macromicro: smaller benchmarks that perform a single task../benchmark/micro
Micro- and macrobenchmarks should ideally terminate on their own, or be easily killable without interacting with the terminal raplrs is being run from.
raplrs outputs data from each poll in csv format.
For each measurement, rows according to the available zones in your CPU are appended.
Sample output:
zone time elapsed (s) power used (joules) avg watt usage avg watt usage since last poll start power (joules) previous power previous power reading
package-0, 28.030436537, 517.9944529999993, 18.479764219175102, 17.634610487045965, 14719.149051, 500.34430299999985, 15237.143504
core, 28.030548686, 364.91492099999596, 13.01855967970924, 11.588462316530928, 55951.045459, 353.3162470000025, 56315.96038
uncore, 28.030660372, 39.756702999999106, 1.4183443342235091, 2.070532459219232, 4492.606319, 37.684346000000005, 4532.363022
A full sample log can be found in ./logs/.
Using isolate as a setup tool, raplrs can estimate pure software energy consumption by offsetting the measurements using previously measured idle data of the system consumption.
What is necessary for this is the output the isolate tool - that is, a JSON on the following format:
[
"package-0": {
"power_j": {
"min": 1.4401819999911822,
"max": 3.130302000005031,
"avg": 1.5812111187977085,
"total": 3315.913507999998
},
"watts": { ... },
"watts_since_last": { ... },
"watt_h": { ... },
"kwatt_h": { ... }
},
"core": { ... },
...
]RAPL.rs 0.1.0
cs-21-pt-9-01
RAPL measurement tool
USAGE:
raplrs [OPTIONS] <SUBCOMMAND>
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
OPTIONS:
-d, --delay <delay> Delay between polls (ms) [default: 1000]
-i, --isolate-from <isolate-file> Idle data to isolate measurements from - see README.md for details
-n, --name <name> Benchmark name - to easily discern csv output
-t, --terminate-after <run-time-limit> Terminate after time limit (s)
SUBCOMMANDS:
benchmark Measure power consumption of a oneshot script
benchmark-int Measure power consumption of an interactive application
help Prints this message or the help of the given subcommand(s)
isolate Tools for measuring and generating isolation data
list List utility for various RAPL-related information
live Live measurements
pretty-print Pretty print last measurement of .csv file
The following system-wide options are available:
-d, --delay: the delay between RAPL polls - that is, the delay between retrieving RAPL measurements from the system in milliseconds. Default is 1.000ms.-n, --name: an identifier for the benchmark, in addition to the mode that is being run - used to name the output.csvfile(s). For example,raplrs -n idle livewould name the.csvfile something likeidle-live-420.69.csv.-t, --terminate-after: the time limit for the benchmark in seconds. For example,raplrs -t 30 livewould terminate the measurements after 30 seconds.-i, --isolate-from: idle data to use to isolate software consumption. This should be generated throughisolate.
Perform live, continuous measurements of power consumption.
raplrs-live 0.1.0
Live measurements
USAGE:
raplrs live [OPTIONS]
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
$ sudo raplrs live
Press 'q' to quit
zone time(s) J avg watt avg watt curr w/h kw/h
package-0 21.01861 401.94149 19.12318 19.61201 0.11165 0.00011
core 21.01864 298.07888 14.18170 13.46358 0.08280 0.00008
uncore 21.01866 20.68407 0.98409 2.15952 0.00575 0.00001
Benchmark a single, oneshot program, optionally n times. If -n is passed, n .csv files will be generated as well.
By default, benchmark expects <program> to be executable - alternatively you can specify a runner, e.g., bash, with -r, --runner.
Additionally, benchmark expects <program> to terminate on its own - if this is not the case for your benchmark, see benchmark-int.
To allow a cooldown period between benchmarks, specify an interval in seconds with -i, --interval.
raplrs-benchmark 0.1.0
Measure power consumption of a oneshot script
USAGE:
raplrs benchmark [OPTIONS] <program> [args]...
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
OPTIONS:
-i, --interval <interval> Interval between benchmark runs in seconds [default: 0]
-n <n> Amount of times to run benchmark [default: 1]
-r, --runner <runner> Benchmark requires <runner> to execute
ARGS:
<program> Benchmark program
<args>... Args for <program>
$ sudo raplrs benchmark benchmark/micro/fib.sh -n 3
Running benchmark iteration 1
# result omitted
Running benchmark iteration 2
# result omitted
Running benchmark iteration 3
zone time(s) J avg watt avg watt curr w/h kw/h
package-0 5.76740 116.93481 20.27543 0.00000 0.03248 0.00003
core 5.76744 89.69783 15.55278 0.00000 0.02492 0.00002
uncore 5.76750 4.55624 0.79001 0.00000 0.00127 0.00000
Benchmark an interactive program.
By default, benchmark-int expects <program> to be executable - alternatively you can specify a runner, e.g., bash, with -r, --runner.
Additionally, benchmark-int will log results in terminal in an ncurses window. To retain availability of the terminal and only log in the background, pass -b, --bg-log.
raplrs-benchmark-int 0.1.0
Measure power consumption of an interactive application
USAGE:
raplrs benchmark-int [FLAGS] [OPTIONS] <program>
FLAGS:
-b, --bg-log Log in background and post a summary on exit
-h, --help Prints help information
-V, --version Prints version information
OPTIONS:
-r, --runner <runner> Benchmark requires <runner> to execute
ARGS:
<program> Benchmark program
$ sudo raplrs benchmark-int benchmark/interactive/cura.sh
Running application "benchmark/interactive/cura.sh"
'q' or ctrl+c to exit. Ctrl+c will kill "benchmark/interactive/cura.sh" as well
zone time(s) J avg watt avg watt curr w/h kw/h
package-0 24.02615 473.20264 19.69537 20.30874 0.13145 0.00013
core 24.02617 346.57723 14.42504 14.78349 0.09627 0.00010
uncore 24.02619 28.36818 1.18072 1.53713 0.00788 0.00001
List utility for various information.
Eligible input:
zones: list all available power zones in CPU
raplrs-list 0.1.0
List utility for various RAPL-related information
USAGE:
raplrs list <input>
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
ARGS:
<input> What to list
$ sudo raplrs list zones
RAPLZone { path: "/sys/devices/virtual/powercap/intel-rapl/intel-rapl:0", name: "package-0" }
RAPLZone { path: "/sys/devices/virtual/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:0", name: "core" }
RAPLZone { path: "/sys/devices/virtual/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1", name: "uncore" }
Pretty print (like the output measurements from measurement tools) the last measurement of a .csv file.
raplrs-pretty-print 0.1.0
Pretty print last measurement of .csv file
USAGE:
raplrs pretty-print <file>
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
ARGS:
<file> File to print from
$ raplrs pretty-print some-log.csv
zone time(s) J avg watt avg watt curr w/h kw/h
package-0 100.09665 2180.14650 21.78043 21.83721 0.60560 0.00061
core 100.09672 1707.28910 17.05642 17.26086 0.47425 0.00047
uncore 100.09676 87.50553 0.87421 0.73465 0.02431 0.00002
Measure or generate isolated system consumption data to be used in benchmarks when -i, --isolate-from is passed.
By default, isolate will measure the system consumption for 30 minutes.
Alternatively, a different value (in minutes) can be specified with -m, --measure.
This data will be logged in a file idle-isolate-420.69.csv - an example can be seen in ./logs/.
To generate the necessary data for isolating the software consumption, pass -g, --generate with the file generated from -m, --measure.
This will create a JSON file containing min, max, avg, and total of the relevant fields used to isolate software consumption.
raplrs-isolate 0.1.0
Tools for measuring and generating isolation data
USAGE:
raplrs isolate [OPTIONS]
FLAGS:
-h, --help Prints help information
-V, --version Prints version information
OPTIONS:
-g, --generate <file> Generate isolation data based on input .csv file
-m, --measure <measure> Measure data as a basis for isolation for n minutes - make sure your system is as idle as
possible [default: 30]
# first, ensure a stable, idle environment
# then measure the system consumption for however long you want with -m <minutes>
# this will output a .csv file as "idle-isolate-STAMP.csv"
$ sudo raplrs isolate -m 30
# then convert the previously measured data to json with relevant information with -g <csv_file>
# this will output a .json file as "isolate-data-STAMP.json"
$ raplrs isolate -g idle-isolate-STAMP.csv
# now use the .json file with the base argument -i, for example
$ sudo raplrs -i isolate-data-STAMP.json live