diff --git a/README.md b/README.md index bd10e51..36c518d 100644 --- a/README.md +++ b/README.md @@ -1,7 +1,6 @@ # gc_spyu -gc_spyu by default obtains node topologies. However, topologies are inaccurate when providers are running from virtualized environments. It is being revised to download topologies as an option instead of by default. +Obtain cpu information from providers and persistently store for later retrieval. Optionally, pry into topology information. -CPU topology is becoming increasingly important in programming to leverage CPUs to maximum efficiency. gc_spyu empowers golem requestors to obtain detailed topology information on any node for mGLM (milli); thereby empowering developers to inspect cache sizes on nodes to enable writing low level code that aligns with the fastest accessible memory banks! For information on the global supercomputer that is Golem and how to run or engineer apps on it, visit https://www.golem.network @@ -11,7 +10,7 @@ For a listing of providers to inspect, visit https://stats.golem.network or invo ```bash $ git clone https://github.com/krunch3r76/gc_spyu.git $ cd gc_spyu -(gc_spyu)$ git checkout v0.0.2 +(gc_spyu)$ git checkout v0.0.3 (gc_spyu)$ python -m venv myvenv (gc_spyu)$ . myvenv/bin/activate (myvenv)$ pip install -r requirements.txt @@ -29,22 +28,25 @@ https://user-images.githubusercontent.com/46289600/152594180-054dad3a-4c53-4103- ``` ## to inspect etam and q53 on testnet, subnet devnet-beta ```bash -(myvenv)$ ./spyu.py --disable-logging --spy etam q53 +(myvenv)$ ./spyu.py --disable-logging --spy etam q53 --add-topology ``` ## to inspect collossus, odra, and whirlwind on testnet, subnet public-beta ```bash -(myvenv)$ ./spyu.py --disable-logging --subnet-tag public-beta --spy collossus odra whirlwind +(myvenv)$ ./spyu.py --disable-logging --subnet-tag public-beta --spy collossus odra whirlwind --add-topology ``` ## to use the gc__filterms / gc__listoffers environment setting ```bash -(myvenv)$ GNPROVIDER=['0x3dd491','0xcef890','etam'] ./spyu +(myvenv)$ GNPROVIDER=['0x3dd491','0xcef890','etam'] ./spyu --add-topology ``` # CREDITS The topology information is taken by running the latest version of openmpi's hwloc. For more information on hwloc, visit: https://www.open-mpi.org/projects/hwloc/. As more CPUs etc topologies emerge, the app is able to update to the latest version by rebuilding the gvmi image from the included Dockerfile. For more on how to rebuild a Golem image, visit https://www.golem.network utils.py from yapapi examples was utilized to start the Golem process and help manage exceptions. To see the original code, browse: https://github.com/golemfactory/yapapi/blob/master/examples/utils.py +## about topology +CPU topology is becoming increasingly important in programming to leverage CPUs to maximum efficiency. gc_spyu empowers golem requestors to obtain detailed topology information on any node for mGLM (milli); thereby empowering developers to inspect cache sizes on nodes to enable writing low level code that aligns with the fastest accessible memory banks! + # MORE Stay tuned, more to come including graphical topologies (not ascii), historical lookups, and gc_listoffers interop (some interop already implied by filterms).