SFC TSS - Service Function Chain (SFC) traffic scheduling simulator - is an Apache2 licensed packet-level discrete event simulator. SFC TSS simulates the SFC traffic scheduling problem as described in our paper "Letting off STEAM: Distributed Runtime Traffic Scheduling for Service Function Chaining". We refer to this paper for more information on RFC 7665 and the features of the simulator. SFC TSS simulates scenarios in compliance with RFC 7665 including link latencies, packet handling at the various SFC components like SFFs, SFIs and server.
If you use SFC TSS in your research, please cite our paper:
@inproceedings{bloecher2020steam,
title={Letting off STEAM: Distributed Runtime Traffic Scheduling for Service Function Chaining},
author={Blöcher, Marcel and Khalili, Ramin and Wang, Lin and Eugster, Patrick},
booktitle={IEEE INFOCOM 2020 - IEEE Conference on Computer Communications},
pages={824-833},
doi = {10.1109/INFOCOM41043.2020.9155404},
year={2020}
}
To setup your environment use either PyPy3 or Python3. We highly recommend you to use PyPy3.
pip install sfctss
Tested with Ubuntu 19.10 / Mac OS 10.15.
SFC TSS provides the essential parts of a SFC traffic simulation and provides many options to configure an experiment.
A minimal configuration requires the following steps
import numpy as np
import random
import sfctss
from sfctss.scheduler.examples import LoadUnawareRoundRobinScheduler
rand = random.Random()
rand.seed(42) # seed the experiment
sim = sfctss.simulator.Sim(seed=rand.randint(0,1000000))
# create a link latency distribution that is used to connect between SFFs-SFIs
LATENCY_SFF_SFI = 1
sfctss.model.SFF.setup_latency_distribution(sim=sim,
id=LATENCY_SFF_SFI,
values=np.random.poisson(500, 5000)) # 3000µs
# create a link latency distribution that is used to connect between SFFs-SFFs
LATENCY_SFF_SFF = 2
sfctss.model.SFF.setup_latency_distribution(sim=sim,
id=LATENCY_SFF_SFF,
values=np.random.poisson(3000, 5000)) # 3000µs
# initialize data structures, configure number of sf types
sfctss.model.SFI.init_data_structure(sim=sim,
number_of_sf_types=1,
latency_provider_sff_sfi=LATENCY_SFF_SFI)
# at least one SFF with a scheduler instance
scheduler_a = LoadUnawareRoundRobinScheduler(sim=sim,
incremental=True, # schedule one step of a chain per scheduling attempt
oracle=True) # scheduler has a global view (all sites)
sff_a = sfctss.model.SFF(sim=sim,
scheduler=scheduler_a)
# at least one Server with a SFI that is connected to the SFF
server = sfctss.model.Server(sim=sim,
processing_cap=120,
cpu_policy=sfctss.model.ServerCpuPolicy.one_at_a_time)
server.add_sfi(of_type=1,
with_sff_id=sff_a.id)
# do the same for ssf_b ...
scheduler_b = None # ...
sff_b = None # ...
# configure connections between SFFs
sfctss.model.SFF.setup_connection(sim=sim,
source_id=sff_a.id,
destination_id=sff_b.id,
bw_cap=100000,
latency_provider=LATENCY_SFF_SFF,
bidirectional=True)
# configure processing speed of sf types
# the rate gives the number of packets a sfi of this sf type can process in 1 s when using 1 cpu share
sfctss.model.SFI.setup_sf_processing_rate_per_1s(sim=sim,
of_type=1,
with_mu=100)
# create at least one packet generator (which could also replay a pcap)
wl_config = sfctss.workload.SyntheticWorkloadGenerator.get_default_config()
wl_gen = sfctss.workload.SyntheticWorkloadGenerator(sim=sim,
workload_rand=rand,
config=wl_config)
sim.register_packet_generator(packet_generator=wl_gen,
fetch_all=False)
# finally, start simulation
sim.run_sim(show_progress=True, # print progress on bash
interactive=False, # no interactive mode
max_sim_time=1500000, # we stop after 1.5s
ui=False, # do not show bash ui
stop_simulation_when_workload_is_over=True) # stop when max_sim_time is done or when workload is done We refer to the example example/main.py for an example how to use SFC TSS.
You may also want to create your own scheduler. Simply subclass BaseScheduler.
Check example schedulers in sfctss.scheduler.examples for more information.
You may also want to create a custom workload provider like a pcap replay. Your workload provider must implement WorkloadGenerator.
Start running the example experiment
./example/main.py --show-progressor with more debugging output or activated statistics dumps
# show progress, write csv logs, active some of the statistics
./example/main.py --show-progres --write-statistics output --statistics-server --statistics-polling-sfi --statistics-latency-cdf-buckets 50
# with more verbose bash ui
./example/main.py --show-ui
# debugging mode
./example/main.py -v --interactive The example provides more options...
usage: main.py [-h] [-v] [--sim-time SIM_TIME] [--no-workload-reloading] [--dry] [--show-progress] [--interactive] [--show-ui]
[--write-statistics STATISTICS_FILENAME] [--statistics-overview] [--statistics-packets] [--statistics-server] [--statistics-polling-sfi]
[--statistics-polling-sff] [--statistics-polling-server] [--statistics-polling-overview]
[--statistics-polling-interval STATISTICS_POLLING_INTERVAL] [--statistics-latency-cdf-buckets STATISTICS_PACKETS_CDF_BUCKETS]
[--dump-full-workload]
optional arguments:
-h, --help show this help message and exit
-v, --verbose verbose output
--sim-time SIM_TIME set simulation time (in ns)
--no-workload-reloading
if set, load full workload before simulation starts
--dry do not run the simulation, but test everything if it is functional
--show-progress shows the progress during running the simulation
--interactive run each simulation tick one after another
--show-ui shows a simple bash-ui when running the simulation
--write-statistics STATISTICS_FILENAME
if filename is set, activate statistics
--statistics-overview
activate overview statistics
--statistics-packets activate packet statistics
--statistics-server activate server statistics
--statistics-polling-sfi
activate sfi polling statistics
--statistics-polling-sff
activate sff polling statistics
--statistics-polling-server
activate server polling statistics
--statistics-polling-overview
activate overview polling statistics
--statistics-polling-interval STATISTICS_POLLING_INTERVAL
set statistics polling interval (in ns)
--statistics-latency-cdf-buckets STATISTICS_PACKETS_CDF_BUCKETS
activate cdf of packet latencies; set # of buckets for cdf, e.g., 50
--dump-full-workload dumps full workload (full packet dump)
Run one of the following lines
./bootstrap-deps-pypy.sh # recommended option
./bootstrap-deps.sh # fallback with standard Pythonto setup your environment either with PyPy3 or Python3. We highly recommend you to use PyPy3.
Simply load your python environment by calling one of the following lines
source env-pypy/bin/activate
source env/bin/activateWe are happy for all kind of contributions, including bug fixes and additional features.
This work has been co-funded by the Federal Ministry of Education and Research (BMBF) Software Campus grant 01IS17050, the German Research Foundation (DFG) as part of the projects B2 and C7 in the Collaborative Research Center (CRC) 1053 “MAKI” and DFG grant 392046569 (61761136014 for NSFC), and the EU H2020 program under grant ICT-815279 “5G-VINNI” and ERC grant FP7-617805 “LiveSoft”.