fluxbind
Intelligent detection and mapping of processors for HPC
Run
Use fluxbind to run a job binding to specific cores. For flux, this means we require exclusive, and then for each node customize the binding exactly as we want it. We do this via a shape file.
Basic Examples
# Start with a first match policy
flux start --config ./examples/config/match-first.toml
# 1. Bind each task to a unique physical core, starting from core:0 (common case)
fluxbind run -n 8 --quiet --shape ./examples/shape/1node/packed-cores-shapefile.yaml sleep 1
# 2. Reverse it!
fluxbind run -n 8 --quiet --shape ./examples/shape/1node/packed-cores-reversed-shapefile.yaml sleep 1
# 3. Packed PUs (hyperthreading), so interleaved.
fluxbind run --tasks-per-core 2 --quiet --shape ./examples/shape/1node/interleaved-shapefile.yaml sleep 1
# 4. Reverse it again!
fluxbind run --tasks-per-core 2 --quiet --shape ./examples/shape/1node/interleaved-reversed-shapefile.yaml sleep 1
# 5. An unbound rank - this tests "unbound" to leave Rank 0 unbound, pack all other ranks onto cores, shifted by one.
fluxbind run -N1 -n 3 --shape ./examples/shape/1node/unbound_rank.yaml sleep 1
# 6. L2 cache affinity. Give each task its own dedicated L2 cache to maximize cache performance.
# On mymachine, each core has its own private L2 cache.
# Therefore, binding one task per L2 cache is equivalent to binding one task per core.
fluxbind run -N1 -n 8 --quiet --shape ./examples/shape/1node/cache-affinity.yaml sleep 1
# 7. Reverse it
fluxbind run -N1 -n 8 --quiet --shape ./examples/shape/1node/cache-reversed-affinity.yaml sleep 1
Kripke Examples
As we prepare to test with apps, here are some tests I'm thinking of doing.
# 1. Baseline - pack each MPI rank onto its own dedicated physical core (8.693519e-09)
fluxbind run -N 1 -n 8 --shape ./examples/shape/kripke/baseline-shapefile.yaml kripke --procs 2,2,2 --zones 16,16,16 --niter 500
# 2. Spread cores (memory bandwidth)
# If Kripke is limited by memory bandwidth, if we place ranks on every other core, we reduce contention for the shared L3 cache
# If Kripke memory bound, this layout might be faster than packed even with half cores. If compute based, worse (1.341355e-08)
fluxbind run -N 1 -n 4 --shape ./examples/shape/kripke/memory-spread-cores-shapefile.yaml kripke --procs 2,2,1 --zones 16,16,16 --niter 500
# 3. Packed pus (each of 8 cores has 2 pu == 16). We are testing if Kripke can benefit from SMT (simultaneous multi-threading)
fluxbind run -N 1 --tasks-per-core 2 --shape ./examples/shape/kripke/packed-pus-shapefile.yaml kripke --procs 2,4,2 --zones 16,16,16 --niter 500
# 4. Hybrid model: launch just two MPI ranks and give each one a whole L3 cache domain to work with (1.966967e-08)
fluxbind run -N 1 -n 2 --env OMP_NUM_THREADS=4 --env OMP_PLACES=cores --shape ./examples/shape/kripke/hybrid-l3-shapefile.yaml kripke --zones 16,16,16 --niter 500 --procs 2,1,1 --layout GZD
Predict
Use fluxbind to predict binding based on a job shape. This is prediction only, meaning there is no execution of an application or similar. Here are some examples.
# Predict binding on this machine for 8 cores
fluxbind predict core:0-7
# Predict binding on corona (based on xml) for 2 NUMA nodes
fluxbind predict --xml ./examples/topology/corona.xml numa:0,1 x core:0-2
License
DevTools is distributed under the terms of the MIT license. All new contributions must be made under this license.
See LICENSE, COPYRIGHT, and NOTICE for details.
SPDX-License-Identifier: (MIT)
LLNL-CODE- 842614
Metadata
Release files for fluxbind 0.0.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| fluxbind-0.0.1.tar.gz | 26.3 kB | Without provenance | Details for fluxbind-0.0.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| fluxbind-0.0.1-py3-none-any.whl | Python 3 | none | any | Details for fluxbind-0.0.1-py3-none-any.whl |
Total release size: 55.0 kB
Release files / fluxbind-0.0.1.tar.gz
| Download URL | fluxbind-0.0.1.tar.gz |
|---|---|
| Size | 26.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
d667c109544d91d6eb3337265800d8168e8ce13e717a4a008bc45573bda9d467
|
|
BLAKE2b-256 checksum How to use checksums |
5f86b99978d29ba66a0b20f16f06856b4c55b744605055d2a55534ccec717687
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/5.1.1 CPython/3.12.2
|
Release files / fluxbind-0.0.1-py3-none-any.whl
| Download URL | fluxbind-0.0.1-py3-none-any.whl |
|---|---|
| Size | 28.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
eeec26b98921e77ec98053be0a14e92a32628039fecf3bf8b87bd3538fed3938
|
|
BLAKE2b-256 checksum How to use checksums |
e57af714a9e6c140cd2ade9f3d80f6a3fd8e167953cddaf83c33fdf971b6773b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/5.1.1 CPython/3.12.2
|
Security for fluxbind 0.0.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.