fiovisualizer

This tool is a realtime data visualisation tool for FIO written using PyQtGraph.

Overview

This application is written in Python, using PyQtGraph to graph FIO data in realtime. Running as root is required because this application accesses block devices.

Package Content

• README.md
  • This readme file
• LICENSE.txt
  • License file
• fio_visualizer.py
  • The frontend for this application
• realtime_back.py
  • The backend for this application
• mainwindow.ui
  • GUI definition part of the script
• Workloads:
  • Device
    • Block device workloads covering typical specifications, separated in SATA and NVMe sections
  • Precondition
    • instruction and script files to prepare SSD for the testing.
  • Simulation
    • Workloads to simulate real applications.
  • MultiDevice
    • Workloads to test two or more drives at a time. Currently has limitted support due to improper handling of total performance in the script.

Setup

Installation

Requires:

• FIO 2.15 or newer (2.1.x is no longer supported) found here
• PyQtGraph found here
• Python 2.7 or 3+ here
• PyQt 4.8+ or PySide
• numpy
• scipy

CentOS 7 steps

0. You should have python and PyQt installed with this distro

1. Install pyqtgraph-develop (0.9.9 required) from here

   • $ python setup.py install

2. Install Cyphon from here Version 0.21 or higher is required.

   • $ python setup.py install

3. Install Numpy from here

   • $ python setup.py build
   • $ python setup.py install

4. Install FIO from here

   • # ./configure
   • # make
   • # make install

5. Run Visualizer as root.

   • # ./fio_visualizer.py

Ubuntu steps

1. Add the following repository deb http://luke.campagnola.me/debian dev/ into your /etc/apt/sources.list and then run $ sudo apt-get install fio python-pyqtgraph

2. Install FIO from here

   • # ./configure
   • # make
   • # make install

3. Run Visualizer as root.

   • # ./fio_visualizer.py

Development notes

Limitations

1. Increased CPU load with numjobs > 4 and all threads plotting. Avoid plotting all threads with numjobs > 4 or disable unneccessary ones to save CPU resources.
2. Multi-jobs configurations files (i.e. [job1]... [job2]) are not supported at the moment. The total amount of jobs if every workload has "numjobs" specified too are not correctly parsed. This can still be implemented manually specifing it in the code.

Todo

Continue working on workload profiles for new Intel SSDs.

Maintainers

• Andrey Kudryavtsev [email protected]
• Alexey Ponomarev [email protected]

Realtime backend processor

fiovisualizer implements its realtime backend processor on the file realtime_back.py. This section documents the backend processor.

Starting fio

The routine start_fio() kicks off the fio process, and it kicks of a process which more or less looks something like the following:

fio --minimal --eta=never --status-interval=1 path-to-fio.ini

Realtime resolution

Since --status-interval=1 is currently used statically, it means a resolution of 1 second is used between updates.

Meaning of data being processed

Since --status-interval is used the values being processed currently do not provide per period measurements. The values being processed reflect the overall cumulative (from job start) values at the specified time intervals.

fiovisualizer support for output format

fio supports different output formats. The smallest and most compact output format currently is the terse output, and so fiovisualizer uses this to minimize the amount of processing done on the output and also to minimize the amount of data being transfered from a remote fio server.

fiovisualizer terse output format support

fio supports different terse format versions, the format version can be specified via the argument --terse-version. If no version is specified the version used will depend on the fio release.

fio terse parsing version notes

When --minimal is used it is equivalent to using --output-format=terse.

The --output-format parameter was added via fio commit f3afa57e3 ("Add --output-format command line option") merged since fio-2.0.10.

fiovisualizer started supporting fio since fio-2.1.14. The release of fio-2.1.14 uses by default the terse output format version 3. fiovisualizer's parser only currently supports terse output format version 3.

The --terse-version parameter was added to fio via commit f57a9c59e36 ("Add terse version output format command line parameter"), this commit was merged on the fio-1.58 release. This fio-1.58 release however used the terse version format 2. Development for the terse version format 3 started since fio commit 312b4af22018a ("Add IOPS to terse output") supported since fio-1.99.6.

The current default terse output format on the latest fio release is terse version format 3.

fio terse output format version 3

Below is a short description of each or the column outputs for fio terse output version format 3.

| 0 | terse_version_3 | | 1 | fio_version | | 2 | jobname | | 3 | groupid | | 4 | error | | 5 | read_kb | | 6 | read_bandwidth | | 7 | read_iops | | 8 | read_runtime_ms | | 9 | read_slat_min | | 10 | read_slat_max | | 11 | read_slat_mean | | 12 | read_slat_dev | | 13 | read_clat_min | | 14 | read_clat_max | | 15 | read_clat_mean | | 16 | read_clat_dev | | 17 | read_clat_pct01 | | 18 | read_clat_pct02 | | 19 | read_clat_pct03 | | 20 | read_clat_pct04 | | 21 | read_clat_pct05 | | 22 | read_clat_pct06 | | 23 | read_clat_pct07 | | 24 | read_clat_pct08 | | 25 | read_clat_pct09 | | 26 | read_clat_pct10 | | 27 | read_clat_pct11 | | 28 | read_clat_pct12 | | 29 | read_clat_pct13 | | 30 | read_clat_pct14 | | 31 | read_clat_pct15 | | 32 | read_clat_pct16 | | 33 | read_clat_pct17 | | 34 | read_clat_pct18 | | 35 | read_clat_pct19 | | 36 | read_clat_pct20 | | 37 | read_tlat_min | | 38 | read_lat_max | | 39 | read_lat_mean | | 40 | read_lat_dev | | 41 | read_bw_min | | 42 | read_bw_max | | 43 | read_bw_agg_pct | | 44 | read_bw_mean | | 45 | read_bw_dev | | 46 | write_kb | | 47 | write_bandwidth | | 48 | write_iops | | 49 | write_runtime_ms | | 50 | write_slat_min | | 51 | write_slat_max | | 52 | write_slat_mean | | 53 | write_slat_dev | | 54 | write_clat_min | | 55 | write_clat_max | | 56 | write_clat_mean | | 57 | write_clat_dev | | 58 | write_clat_pct01 | | 59 | write_clat_pct02 | | 60 | write_clat_pct03 | | 61 | write_clat_pct04 | | 62 | write_clat_pct05 | | 63 | write_clat_pct06 | | 64 | write_clat_pct07 | | 65 | write_clat_pct08 | | 66 | write_clat_pct09 | | 67 | write_clat_pct10 | | 68 | write_clat_pct11 | | 69 | write_clat_pct12 | | 70 | write_clat_pct13 | | 71 | write_clat_pct14 | | 72 | write_clat_pct15 | | 73 | write_clat_pct16 | | 74 | write_clat_pct17 | | 75 | write_clat_pct18 | | 76 | write_clat_pct19 | | 77 | write_clat_pct20 | | 78 | write_tlat_min | | 79 | write_lat_max | | 80 | write_lat_mean | | 81 | write_lat_dev | | 82 | write_bw_min | | 83 | write_bw_max | | 84 | write_bw_agg_pct | | 85 | write_bw_mean | | 86 | write_bw_dev | | 87 | cpu_user | | 88 | cpu_sys | | 89 | cpu_csw | | 90 | cpu_mjf | | 91 | cpu_minf | | 92 | iodepth_1 | | 93 | iodepth_2 | | 94 | iodepth_4 | | 95 | iodepth_8 | | 96 | iodepth_16 | | 97 | iodepth_32 | | 98 | iodepth_64 | | 99 | lat_2us | | 100 | lat_4us | | 101 | lat_10us | | 102 | lat_20us | | 103 | lat_50us | | 104 | lat_100us | | 105 | lat_250us | | 106 | lat_500us | | 107 | lat_750us | | 108 | lat_1000us | | 109 | lat_2ms | | 110 | lat_4ms | | 111 | lat_10ms | | 112 | lat_20ms | | 113 | lat_50ms | | 114 | lat_100ms | | 115 | lat_250ms | | 116 | lat_500ms | | 117 | lat_750ms | | 118 | lat_1000ms | | 119 | lat_2000ms | | 120 | lat_over_2000ms | | 121 | disk_name | | 122 | disk_read_iops | | 123 | disk_write_iops | | 124 | disk_read_merges | | 125 | disk_write_merges | | 126 | disk_read_ticks | | 127 | write_ticks | | 128 | disk_queue_time | | 129 | disk_util |

Processing

Below is an example output of one line out using:

fio  --client=some-hostname --eta=never --status-interval=1 --terse-version=3 some.ini

The output:

3;fio-3.8-41-g7302;job1;0;0;0;0;0;0;0;0;0.000000;0.000000;0;0;0.000000;0.000000;1.000000%=0;5.000000%=0;10.000000%=0;20.000000%=0;30.000000%=0;40.000000%=0;50.000000%=0;60.000000%=0;70.000000%=0;80.000000%=0;90.000000%=0;95.000000%=0;99.000000%=0;99.500000%=0;99.900000%=0;99.950000%=0;99.990000%=0;0%=0;0%=0;0%=0;0;0;0.000000;0.000000;0;0;0.000000%;0.000000;0.000000;1886832;188645;23580;10002;1;87;2.561636;0.999742;6;6575;165.280554;282.700285;1.000000%=11;5.000000%=13;10.000000%=15;20.000000%=19;30.000000%=25;40.000000%=35;50.000000%=51;60.000000%=89;70.000000%=148;80.000000%=240;90.000000%=452;95.000000%=675;99.000000%=1286;99.500000%=1679;99.900000%=2736;99.950000%=3293;99.990000%=5013;0%=0;0%=0;0%=0;12;6577;167.875769;282.729144;138363;314224;25.026987%;188630.900000;35637.687291;5.179482%;9.399060%;151970;0;51;0.1%;0.1%;100.0%;0.0%;0.0%;0.0%;0.0%;0.00%;0.00%;0.01%;21.49%;27.82%;12.61%;18.75%;10.68%;4.67%;2.00%;1.66%;0.28%;0.03%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;nvme0n1;44;934217;0;0;9;152472;152997;99.24%

From the above fiovisualizer's realtime_back.py records the following 6 columns and graphs them, below listed from the respective value from the output above:

| 6 | read_bandwidth | 0 | | 7 | read_iops | 0 | | 15 | read_clat_mean | 0.000000 | | 47 | write_bandwidth | | 188645 | | 48 | write_iops | 23580 | | 56 | write_clat_mean | 165.280554 |