Coder Social home page Coder Social logo

pwrapi-ref's Introduction

This software is a reference implementation of the PowerAPI. The main purpose 
for the creation of the reference was to test the viability of API functions
during the design of the API specification. The reference is also being used
to test implementation details such as how to interface with different types
of hardware and communication between agents. The reference is a work in
progress and only a subset of the functions listed in the specification have
been implemented.

The main component of this software is libpwr (located in directory pwr). 
Any application that uses the PowerAPI must link against libpwr.so. 
It implements the API and is designed as a hardware neutral framework. The
framework accesses platform specific power measurement and control via plugins.
The plugins are implemented as dynamic libraries (located in directory plugins).
There are currently plugins for several platforms as well as a dummy plugin
used for development. The dummy plugin is trivial and a good place to look when
implementing a plugin for a new platform. 

libpwr can be configured via a static XML file describing the hardware or via
hwloc. XML was chosen for expedient development not performance. libpwr has an
object that interfaces with the XML file via TinyXML-2. TinyXML-2 is 
included with this distribution and is already integrated into this
software package.  The original software can also be found at 
"https://github.com/leethomason/tinyxml2". Creating a XML configuration is
error prone so support for automatic hierarchy detection was added to the
implementation. This automatic method is called hwloc because it uses hwloc to
determine the Power object hierarchy. 

The reference also explores accessing the API via a scripting language such as
Python. The Python binding is done via Swig and is
located in directory swig. 

The examples directory contains simple โ€œCโ€ and Python programs that can be 
examined to get a feeling for how one would use the Power API. Note that the
execution of example programs rely on environment variables POWERAPI_CONFIG and
POWERAPI_ROOT. POWERAPI_CONFIG tells the PowerAPI library what system
configuration file to use. POWERAPI_ROOT tells the PowerAPI library its entry
point. Consult the run scripts in examples on how to set these variables.

Build
-----

The configuration of this code base follows normal autoconf/automake conventions. 

run:

autogen.sh
configure --prefix="your install directory"
make install

Your install directory should look like:

volta:~/pwrGIT/pwrapi-ref/build/install> ls
bin  config  examples  include  lib  lib64  script  share
volta:~/pwrGIT/pwrapi-ref/build/install>

Run
---

With the above configuration the only thing that will work is the "dummyTest"
program. The "dummyTest" program is used to test basic functionality of the
implementation and as an example of how to program using the Power API. The "dummyTest" 
example uses xml configuration and the dummy_dev plugin. The dummy_dev plugin is just
a stub and does not interact with hardware.

To run "dummyTest".

volta:~/pwrGIT/pwrapi-ref/build/install> source examples/script/xml_profile
volta:~/pwrGIT/pwrapi-ref/build/install> dummyTest
I am a `Node`
child plat.cab0.node0.core0
child plat.cab0.node0.core1
PWR_ObjAttrGetValue(PWR_ATTR_ENERGY) value=200000000.000000 ts=`Fri Apr 29 13:08:11 2016`
PWR_ObjAttrSetValue(PWR_ATTR_ENERGY) value=25.812000
PWR_ObjAttrGetValues(PWR_ATTR_ENERGY) value=51.624000 ts=`Fri Apr 29 13:08:11 2016`
PWR_ObjAttrSetValues(PWR_ATTR_ENERGY) value=100.100000
PWR_ObjAttrGetValue(PWR_ATTR_ENERGY) value=200.200000 ts=`Fri Apr 29 13:08:11 2016`
PWR_GrpAttrSetValue(PWR_ATTR_ENERGY) value=0.100000
PWR_ObjAttrGetValue(PWR_ATTR_ENERGY) value=0.200000 ts=`Fri Apr 29 13:08:11 2016`
PWR_StatGetValue(PWR_ATTR_ENERGY) value=101.081396
PWR_StatGetValue(PWR_ATTR_ENERGY) start=1461956891.332830
PWR_StatGetValue(PWR_ATTR_ENERGY) stop=1461956892.332830
stat: value=101.129473 start=1461956892.332965 stop=1461956893.332965
stat: value=100.870015 start=1461956892.332970 stop=1461956893.332970
volta:~/pwrGIT/pwrapi-ref/build/install>


Running with Hwloc
------------------

If you have "hwloc" installed on your system you can configure via:

configure --prefix="your install directory" --with-hwloc="where hwloc is installed" 

This configuration adds the ability to auto detect the Power object hierarchy
as mentioned above. 

Once built you can run "pwrls". "pwrls" traverses the Power object hierarchy and prints the
objects and supported attributes. Note that dummy_dev plugin is used.

volta:~/pwrGIT/pwrapi-ref/build/install> source script/hwloc_profile
volta:~/pwrGIT/pwrapi-ref/build/install> pwrls
name=`plat` type=Platform: Energy
    name=`plat.node0` type=Node: Energy
        name=`plat.node0.socket0` type=Socket: Energy
            name=`plat.node0.socket0.core0` type=Core: Power Energy
            name=`plat.node0.socket0.core1` type=Core: Power Energy
            name=`plat.node0.socket0.core2` type=Core: Power Energy
            name=`plat.node0.socket0.core3` type=Core: Power Energy
            name=`plat.node0.socket0.core4` type=Core: Power Energy
            name=`plat.node0.socket0.core5` type=Core: Power Energy
            name=`plat.node0.socket0.core6` type=Core: Power Energy
            name=`plat.node0.socket0.core7` type=Core: Power Energy
volta:~/pwrGIT/pwrapi-ref/build/install>

One thing to note about the above output is that some objects have both Power and Energy 
attributes and some have only Energy. This is because Energy can be aggregated from children and Power can not.


Running with the Linux Power Capping Framework Plugin
------------------------------------------------------

Plugin libpwr_powercap.a is a proof of concept plugin that allows a user
application to read energy data through the Linux Power Capping Framework.

To use this plugin with hwloc configuration run:

source install/script/hwloc_powercap_profile 

To see what the Power API hierarchy looks like run:

[mjleven@node69 script]$ pwrls
name=`plat` type=Platform: Energy
    name=`plat.node0` type=Node: Energy
        name=`plat.node0.mem0` type=Memory: Energy
        name=`plat.node0.mem1` type=Memory: Energy
        name=`plat.node0.socket0` type=Socket: Energy
        name=`plat.node0.socket1` type=Socket: Energy
[mjleven@node69 script]$

The above was run on a dual socket Intel(R) Xeon(R) CPU E5-2698 v3 @ 2.30GHz 
machine.

Now lets read the energy:

[mjleven@node69 script]$ pwrenergy -o plat.node0.socket0 -s 5
plat.node0.socket0: 22 joules, 0 watts
plat.node0.socket0: 46 joules, 24 watts
plat.node0.socket0: 67 joules, 21 watts
plat.node0.socket0: 79 joules, 12 watts
plat.node0.socket0: 92 joules, 12 watts
[mjleven@node69 script]$

[mjleven@node69 script]$ pwrenergy -o plat.node0.mem0 -s 5
plat.node0.mem0: 15 joules, 0 watts
plat.node0.mem0: 29 joules, 15 watts
plat.node0.mem0: 44 joules, 15 watts
plat.node0.mem0: 59 joules, 15 watts
plat.node0.mem0: 73 joules, 15 watts
[mjleven@node69 script]$

[mjleven@node69 script]$ pwrenergy -o plat.node0 -s 4
plat.node0: 49 joules, 0 watts
plat.node0: 99 joules, 49 watts
plat.node0: 148 joules, 49 watts
plat.node0: 198 joules, 50 watts
[mjleven@node69 script]$

Note that the powercap plugin does not check for counter wrap around.
Note that you need a linux kernel that supports the "Power Capping Framework".
Note that this plugin has minimal testing on one machine.

Running with the RAPL Plugin
----------------------------

Plugin libpwr_rapldev reads energy data via /dev/cpu/*/msr. You must have
elevated privilege to use this plugin.
 
[root@gato2 script]# source hwloc_rapl_profile
[root@gato2 script]# pwrls
name=`plat` type=Platform: Energy
    name=`plat.node0` type=Node: Energy
        name=`plat.node0.mem0` type=Memory: Energy
        name=`plat.node0.mem1` type=Memory: Energy
        name=`plat.node0.socket0` type=Socket: Energy
        name=`plat.node0.socket1` type=Socket: Energy
[root@gato2 script]# pwrenergy -o plat.node0
plat.node0: 314996 joules, 0 watts
[root@gato2 script]# pwrenergy -o plat.node0 -s 5
plat.node0: 315257 joules, 0 watts
plat.node0: 315296 joules, 39 watts
plat.node0: 315334 joules, 38 watts
plat.node0: 315372 joules, 38 watts
plat.node0: 315410 joules, 38 watts
[root@gato2 script]# pwrenergy -o plat.node0.mem0 -s 5
plat.node0.mem0: 42660 joules, 0 watts
plat.node0.mem0: 42668 joules, 8 watts
plat.node0.mem0: 42677 joules, 8 watts
plat.node0.mem0: 42685 joules, 8 watts
plat.node0.mem0: 42693 joules, 8 watts
[root@gato2 script]# pwrenergy -o plat.node0.socket0 -s 5
plat.node0.socket0: 107759 joules, 0 watts
plat.node0.socket0: 107772 joules, 13 watts
plat.node0.socket0: 107785 joules, 13 watts
plat.node0.socket0: 107797 joules, 13 watts
plat.node0.socket0: 107810 joules, 13 watts
[root@gato2 script]#

Note that the energy value for the socket object is derived from adding 
PP0 and PP1. The energy value for the memory object is based on RAPL DRAM. 

Debug
-----

To debug configure with --enable-debug. At this point the debug 
code is compiled into the libraries but will not print. To enable printing you must set the 
environment variable POWERAPI_DEBUG=31. This will turn on all debug.


If you have questions or comments email [email protected].

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.