nfv-t-cp: NFV Time-Constrained Profiler

NFV time-constrained performance profiling (T-CP) framework.

This framework can be used to build NFV profiling systems that profile single VNFs or complex SFC under a given time constraint. The framework can be either connected to a real-world profiling platform or it can be fed with existing profiling results to perform trace-based T-CP runs independently of a connected profiling platform.

Contact

Manuel Peuster
Paderborn University
manuel (dot) peuster (at) upb (dot) de
Twitter: @ManuelPeuster
GitHub: @mpeuster

Datasets

The datasets used by and generated with this tool are publicly available. If you use them, please cite the following papers:

[Peuster2017] M. Peuster and H. Karl: Profile Your Chains, Not Functions: Automated Network Service Profiling in DevOps Environments. IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN), Berlin, Germany. (2017)
[Peuster2018] M. Peuster, H. Karl: Understand your chains and keep your deadlines: Introducing time-constrained profiling for NFV. In IEEE/IFIP 14th International Conference on Network and Service Management (CNSM), Rome, Italy. (2018)

The results are Pandas data frames exported as CSV files and are available in this Amazon Cloud folder.

peuster_karl_ieeenfvsdn17_3vnf_sfc_profiling_measurements_clean.csv (100.2 KB): Cleaned up RAW measurements of a SFC with three layer 4 forwarding VNFs as presented in [Peuster2017].
2018-04-26-experiment_tc_paper_synthetic_all.combined.csv (5.81 GB!): T-CP system evaluation (e.g. different selector and predictor algorithms) using our randomized synthetic SFC-UT performance model.
2018-04-26-experiment_tc_paper_nfvsdn17_measurements.combined.csv (6.00 GB!): T-CP system evaluation (e.g. different selector and predictor algorithms) using real-world SFC performance measurements.

Installation

python3 setup.py install

Run (single job)

nfvtcp -c example_experiment.yaml

Run (parallel jobs)

./nfvtcp_parallel.sh -c example_experiment.yaml -J 16

The argument -J specifies the number of parallel jobs to use.

Combine results of parallel jobs

If the data frames of multiple parallel jobs should be combined do:

cd out/
python combine_results.py -i "experiment_*.pkl" -o combined_result.pkl

NOTE: The quotes for the file pattern are important!

Build-in plotting

nfvtcp -c example_experiment.yaml --plot out/result.pkl

Tests

To run nfv-t-cp's unit tests do:

pytest
pytest -v -s --log-level DEBUG # be more verbose
pytest -v -s -k "midpoint" # run tests that match keyword

Examples

# run
./nfvtcp_parallel.sh -J 16 -c configs/experiment_tc_paper_synthetic_all.yaml -r 100
./nfvtcp_parallel.sh -J 16 -c configs/experiment_tc_paper_nfvsdn17_measurements.yaml -r 100

# process results
python combine_results.py -i "experiment_tc_paper_synthetic_all.job*" -o 2018-04-XX-experiment_tc_paper_synthetic_all.compressed.combined.pkl
python combine_results.py -i "experiment_tc_paper_nfvsdn17_measurements.job*" -o 2018-04-XX-experiment_tc_paper_nfvsdn17_measurements.combined.compressed.pkl

Contributor(s)

Manuel Peuster (manuel (at) peuster (dot) de)
...

License

Apache 2.0

CN-UPB / nfv-t-cp