Closed cmcantalupo closed 6 years ago
cmcantalupo
commented
6 years ago The GEOPM team has just minted release v0.4.0 of GEOPM https://github.com/geopm/geopm/releases/tag/v0.4.0. Next week I will work on updating the OpenHPC submission and fork of the OpenHPC repo to reference the latest release.
cmcantalupo
commented
6 years ago Submission, OpenHPC fork, and OBS builds have been updated for GEOPM version 0.4.0.
koomie
commented
6 years ago Thank you for the submission and builds under OBS. The TSC has recommended acceptance of GEOPM.
cmcantalupo
commented
6 years ago Great news. Thanks!
Software Name
Global Extensible Open Power Manager (GEOPM)
Public URL
http://geopm.github.io/geopm/
Technical Overview
Global Extensible Open Power Manager (GEOPM) is an extensible power management framework targeting high performance computing. The library can be extended to support new control algorithms and new hardware power management features. The GEOPM package provides built-in features ranging from static management of power policy for each individual compute node, to dynamic coordination of power policy and performance across all of the compute nodes hosting one MPI job on a portion of a distributed computing system. The dynamic coordination is implemented as a hierarchical control system for scalable communication and decentralized control. The hierarchical control system can optimize for various objective functions including maximizing global application performance within a power bound. The root of the control hierarchy tree can communicate through shared memory with the system resource management daemon to extend the hierarchy above the individual MPI job level and enable management of system power resources for multiple MPI jobs and multiple users by the system resource manager. The geopm package provides the libgeopm library, the libgeopmpolicy library, the geopmctl application and the geopmpolicy application. The libgeopm library can be called within MPI applications to enable application feedback for informing the control decisions. If modification of the target application is not desired then the geopmctl application can be run concurrently with the target application. In this case, target application feedback is inferred by querying the hardware through Model Specific Registers (MSRs) or other hardware interfaces. With either method (libgeopm or geopmctl), the control hierarchy tree writes processor power policy through hardware interfaces to enact policy decisions. The libgeopmpolicy library is used by a resource manager to set energy policy control parameters for MPI jobs. Some features of libgeopmpolicy are available through the geopmpolicy application including support for static control.
The GEOPM package provides new functionality to the OpenHPC software stack through its runtime job power management. In order to enable optimal power management, the GEOPM runtime also provides a significant performance profiling framework which complements some of the existing performance profiling tools that are provided by OpenHPC. A significant difference between GEOPM and the other profiling tools is that GEOPM runs on an isolated CPU, and other than removing a CPU from the pool of available to the application, it has a very low impact on the performance of the application due to its remote sampling design.
The GEOPM development effort is a partnership between developers and researchers throughout the HPC community. The GEOPM project is in large part funded by an ongoing NRE grant from Argonne National Lab and we collaborate with power researchers there to provide a path to an exascale system with advanced power management. We partner with the Sandia PowerAPI group to ensure that the GEOPM interfaces are compatible with the interfaces defined by the PowerAPI for a power aware runtime. We also partner with the Lawrence Livermore National Lab power researchers to develop new power management algorithms and extend research ideas developed there into production software. Additionally we have a partnership with the Leibniz Supercomputing Centre to extend the research they have done into providing an at-scale supercomputer while meeting strict energy efficiency goals. We have partnered with researchers at IBM and the Hartree Centre to port GEOPM to the IBM Power architecture.
Latest stable version number
v0.4.0
Open-source license type
BSD 3-clause
Relationship to component?
If other, please describe:
Build system
If other, please describe:
Does the current build system support staged path installations? For example:
make install DESTIR=/tmp/foo(or equivalent)Does component run in user space or are administrative credentials required?
Does component require post-installation configuration.
If yes, please describe briefly:
If component is selected, are you willing and able to collaborate with OpenHPC maintainers during the integration process?
A fork of the OpenHPC repository that includes the GEOPM component can be found here:
https://github.com/cmcantalupo/ohpc/tree/cmcantalupo-geopm
and the builds in the OpenHPC build server can be found here:
https://build.openhpc.community/project/show/home:cmcantalupo
Does the component include test collateral (e.g. regression/verification tests) in the publicly shipped source?
If yes, please briefly describe the intent and location of the tests.
The shipped source code includes unit tests based on Google Test framework as well as integration tests build on the python py-unit test framework. Additionally GEOPM provides a tutorial suite which can provide a basic smoke test.
The unit tests are here:
https://github.com/geopm/geopm/tree/v0.4.0/test
The intent of the unit tests is to test each class independently and quickly with minimal system software and hardware requirements. These tests can be executed in a simple virtual machine as is done by our Continuous Integration process in Travis and the central OpenSUSE Build Server for every commit integrated into our repository.
The integration tests are here:
https://github.com/geopm/geopm/tree/v0.4.0/test_integration
The intent of the integration tests is to prove that the features that GEOPM provides can be shown to be valid. These tests bring up the full GEOPM runtime and execute a configurable synthetic application under the control of the runtime. The tests show that our report and trace capabilities give consistent data, and that the goals of our decision algorithms are met. These tests are run nightly on our test cluster.
And the tutorials are here:
https://github.com/geopm/geopm/tree/v0.4.0/tutorial
The tutorials are intended to give a new user of the GEOPM software an introduction to the features it provides in a step by step process. The build of the tutorials is based on a very simple Makefile system and is independent of the rest of the GEOPM build infrastructure. We have integrated the build and execution of one of the tutorials as our OpenHPC install smoke test.
Does the component have additional software dependencies (beyond compilers/MPI) that are not part of standard Linux distributions?
If yes, please list the dependencies and associated licenses.
There are several dependencies that can be fulfilled by packages maintained in standard Linux distributions, however one dependency not part of the standard Linux distribution is the msr-safe Linux kernel driver developed by Lawrence Livermore National Lab:
https://github.com/LLNL/msr-safe
The msr-safe kernel driver is commonly used at a significant portion of the major HPC facilities. These include the Theta supercomputer at Argonne, the DOE systems running TOSS including facilities at LLNL, LANL and Sandia, at the LRZ SuperMUC-2 system and experimentally on the NERSC Cori system. The likwid software package is another user of the msr-safe kernel driver which recently submitted for inclusion in the OpenHPC distribution.
The msr-safe driver targets HPC systems as it makes two assumptions that are not compatible with the most general use cases of the Linux operating system. The first assumption is compute node single occupancy: at any single point in time, only one user has access to a compute node. The second assumption is that a script can be executed at the time the compute resources are provided to a new user, and after the compute resources are reclaimed from that user. Given these assumptions, which are commonly met in the HPC environment, the msr-safe driver safely enables the types of hardware measurement and control required by GEOPM with an overhead low enough to meet the performance requirements of the GEOPM runtime. The msrsave utility provided with the msr-safe driver enables save/restore functionality to isolate user access between allocations.
Does the component include online or installable documentation?
If available online, please provide URL.
The GEOPM man pages are published here and are also installed as part of the GEOPM build (see SEE ALSO section for links to the other GEOPM man pages):
http://geopm.github.io/geopm/man/geopm.7.html
The GEOPM developer doxygen pages are here:
http://geopm.github.io/geopm/dox/index.html
The GEOPM plugin developer guide is here:
http://geopm.github.io/geopm/pdf/geopm-plugin-developer-guide.pdf
[Optional]: Would you like to receive additional review feedback by email?