Releases: SX-Aurora/py-veosinfo
veperf with Aurora VE3 support for power measurement
The power measurement for Aurora Vector Engine VE3 has been fixed to read the proper sensors, the power consumption of each VE has now an offset of 5 Watts taking into consideration the power consumption of the VMC Arm chip and some more unmeasured bits.
The veperf utility now requires python3!
The build process was streamlined, it is done inside a venv, where needed. You must have python3-devel and veosinfo-devel installed in order to build from the repository.
numa_info, bandwidths, power and energy in veperf
Added numa_info() function.
veperf can now print memory bandwidth information from running processes if they are run with VE_PERF_MODE=VECTOR-MEM.
In addition it prints the power consumption of each VE as well as the energy consumed since the start of the veperf program.
If you have no root privileges on the machine, untar the file py-veosinfo-2.5.1-localinstall-py27-py36.tgz in your home directory. You can then call
~/.local/bin/veperf
calc_metrics in veosinfo package
Moved calc_metrics() function into a pure python source inside the veosinfo package. It can now be used by other components that acquire metrics with the ve_pid_perf() function.
py-veosinfo v1.3.2a incl. sched_*affinity()
This release includes sched_getaffinity() and sched_setaffinity() for VEs.
It is installable (also into virtualenvs) from PYPI:
pip install py-veosinfo
v1.3.2-2 py-veosinfo and veperf
Release which includes a better README-md which documents almost all implemented functions.
The build system is changed such that packages can now be uploaded to PyPI, and installed from there with
pip install py-veosinfo
veperf VE monitoring
Added functions for task metrics measurement.
A goodie is veperf, a performance monitoring tool for the own VE tasks. Check out this post.