bench 2.8

Benchmark resources usage

Monitor Process Resources Usage

Bench aims to be a handy tool with these functions:

- Monitor CPU time (user time, system time, real time)

- Monitor memory usage (virtual memory usage, resident memory usage)

- Output to TSV(tab-delimited files)

- Output benchmark metrics

- Visualize performance metrics (memory consumptions) over time

Examples

We showed several examples below. Please note that all output are tabularized for demonstration purpose.

• Example 1: simple command.

This will start the process sleep for 2 seconds. The tabular output below was from the actual command: monitor.py sleep 2 2>&1 |column -t -s $'\t'.

If you simply run monitor.py sleep 2, you will get tab-deliminated outputs in standard error (stderr).

$> monitor.py sleep 2
pid     ppid    utime  stime  rtime         rss     vms      maxRss  maxVms   avgRss    avgVms     cwd                                cmd
133692  133675  0.0    0.0    1.9368159771  774144  6066176  774144  6066176  774144.0  6066176.0  /home/zhanxw/mycode/bench/scripts  sleep 2

• Example 2: complex shell commands with sampling interval equaling to 0.1 second

This example will use shell to start 3 processes: sleep 2, sleep 4 and seq 1000000. You can see bench can monitor all 4 processes all together.

$> monitor.py sh -c 'sleep 2 & sleep 4 & seq 1000000 >/dev/null & wait'
pid     ppid    utime  stime  rtime            rss     vms      maxRss  maxVms   avgRss    avgVms     cwd                                cmd
135004  134985  0.0    0.0    3.9532430172     798720  4558848  798720  4558848  798720.0  4558848.0  /home/zhanxw/mycode/bench/scripts  sh -c sleep 2 & sleep 4 & seq 10000000 >/dev/null & wait
135006  135004  0.0    0.0    3.95348381996    655360  6066176  655360  6066176  655360.0  6066176.0  /home/zhanxw/mycode/bench/scripts  sleep 4
135005  135004  0.0    0.0    1.83160495758    774144  6066176  774144  6066176  774144.0  6066176.0  /home/zhanxw/mycode/bench/scripts  sleep 2
135007  135004  0.05   0.0    0.0599648952484  720896  6090752  720896  6090752  720896.0  6090752.0  /home/zhanxw/mycode/bench/scripts  seq 10000000

• Example 3: generate performance metrics to external file

Here we used a small program, burnCpu. It will keep CPU running for several seconds. Its source code is under src/.

The option -t will enable outputting traces. That means at several time stops, performance metrics of each processes will be outputted to the standard error as well as a separate comma-separated file, $prefix.trace.csv.

The option -g will generate a graph which contains several sub-figures, including timings for each processes, memory consumption for each processes, and memory consumption over the processing running time.

The option -o will specify the output prefix. The default value will be bench, meaning, you will get bench.csv. You can overwrite this value by using -o option.

$> monitor.py -t -g -o burnCpu ./burnCpu
pid     ppid    utime  stime  rtime            rss      vms       cwd                                cmd
135471  135454  0.04   0.0    0.0441780090332  1449984  12984320  /home/zhanxw/mycode/bench/scripts  ../src/burnCpu
135471  135454  0.2    0.0    0.205282926559   1449984  12984320  /home/zhanxw/mycode/bench/scripts  ../src/burnCpu
135471  135454  0.38   0.0    0.381079912186   1449984  12984320  /home/zhanxw/mycode/bench/scripts  ../src/burnCpu
...

Additional result are stored in burnCpu.csv, burnCpu.trace.csv in the Comma-separated format (CSV).

burnCpu.csv file content

pid,ppid,utime,stime,rtime,rss,vms,maxRss,maxVms,avgRss,avgVms,cwd,cmd
144433,144416,5.4,0.0,5.40555810928,1404928,12984320,1404928,12984320,1404928.0,12984320.0,/home/zhanxw/mycode/bench/scripts,../src/burnCpu

burnCpu.trace.csv file content

pid,ppid,utime,stime,rtime,rss,vms,cwd,cmd
144433,144416,0.03,0.0,0.0423669815063,1404928,12984320,/home/zhanxw/mycode/bench/scripts,../src/burnCpu
144433,144416,0.19,0.0,0.20046210289,1404928,12984320,/home/zhanxw/mycode/bench/scripts,../src/burnCpu
144433,144416,0.36,0.0,0.373480081558,1404928,12984320,/home/zhanxw/mycode/bench/scripts,../src/burnCpu
...

When -g optioned is specified, bench will generate several performance metrics in the file burnCpu.trace.csv:

Notes

To benchmark a complex command or combinations of commands, you can use shell (sh or bash) . For example, you can use “sh -c ‘command arg1 arg2 … ‘” (see Example 2).

Bench requires psutil to collect basic performance metrics, and

requires numpy and pandas for statistical calculations.

In this release, we used psutil 3.1.1, numpy 1.8.2, pandas 0.16.2 and matplotlib 1.4.3.

Contact

For questions or commend, please visit bench github repo:

repo

or email to:

Xiaowei Zhan