This is a simulation of a 3G MVNO virtual mobile core on protocol level including a basic device model as well as detailed interactions of different mobile core components.
This repository accompanies the following publication:
Geißler, S., Wamser, F., Bauer, W., Gebert, S., Kounev, S., & Hoßfeld, T. (2023). MVNOCoreSim: A Digital Twin for Virtualized IoT-centric Mobile Core Networks. IEEE Internet of Things Journal.
The simulation consists of two distinct models, the signaling model and the MVNO core model, as shown in the figure above.
Devices, represented by the signaling interaction between VLR/SGSN and HLR, aim to complete their attachment cycle consisting of the four depicted signaling messages, while attempting a retry if the timers are triggered.
The core model processes arriving signaling messages and transmits responses to the issuing device.
Note that we call these objects devices, although we are not nececarrily talking about physical IoT devices. Instead, devices in our simulation are abstract elements that generate signaling load.
The simulation has been developed using the r-simmer discrete event simulation framework for the R programming language.
To run the example provided in this repository (run_sim.R), the following packages need to be present in the R environment:
- simmer
- tidyverse -- Actually only ggplot2, margrittr and dplyr
- data.table -- For longer running simulation to keep evaluation performant
- futile.logger -- To provide output of the simulation progress
install.packages(c("simmer", "tidyverse", "data.table", "futile.logger"))The code is structured as follows
.
├── LICENSE
├── README.md
├── input # Contains the measured processing time samples for each of the code components
│ └── processing_times.csv
├── resources # Contains the overview image and other resources
│ └── sim_schematic.pdf
├── run_sim.R # Main file to run
└── sim # Contains the actual simulation source
├── sample_helpers.R # Helper files to sample the processing times
├── sim_baseline_recovery.R # Main loops for the different overload control mechanisms
├── sim_device_policing_recovery.R # ...
├── sim_msu_policing_recovery.R # ...
├── sim_simple_dropping_recovery.R # ...
└── trajectories # Contains the trajectory definitions for each of the different mechanisms
├── baseline_recovery
│ ├── devices.R # Defines how devices behave
│ ├── helpers.R # Defines helper trajectories
│ ├── resources.R # Defines component behavior, including processing times
│ ├── sai.R # Defines trajectories for sendAuthenticationInfo dialogs
│ ├── sai_unknown_subscriber.R # Currently unused
│ ├── ul.R # Defines trajectories for updateLocation dialogs
│ ├── ul_gprs.R # Defines trajectories for updateGprsLocation dialogs
│ ├── ul_gprs_roaming_not_allowed.R # Currently unused
│ ├── ul_gprs_unknown_subscriber.R # Currently unused
│ └── ul_roaming_not_allowed.R # Currently unused
├── device_policing_recovery
│ ├── ...
├── msu_policing_recovery
│ ├── ...
├── simple_dropping_recovery
│ ├── ...The run_sim.R main file includes a preconfigured example simulation using the baseline version without dedicated overload control mechanisms.
The simulation parameters are defined as below.
config <- list(
n_devices = 50000, # The number of devices to simulate
init_rate = 300, # The arrival rate of new devices
runtime = 10, # Max runtime of the simulation - mostly relevant for overload scenarios in which consinuously fail to attach to the system
policing = list(
# Parameters for MSU Policing
sai_vlr = 3000, # Nr. of concurrently allowed dialogs for each of the types
sai_sgsn = 3000,
ul = 3000,
ul_gprs = 3000,
# Parameters for Device Policing
n_dev = 3000 # Nr. of concurrently allowed devices to have any dialog in the system
),
capacity = list( # CPU count for each of the core components
cmap = 8,
hlr = 64,
spc = 128,
auc = 8,
isd = Inf # ISD is a special "dummy" component representing the external network, therefore Infinite capacity is recommended
),
queue = list( # Queue size of each of the components
cmap = Inf,
hlr = Inf,
spc = Inf,
auc = Inf,
isd = Inf
),
retry = list( # Time each device waits before reset and retry in seconds after the respective dialog has been issued
sai = 6,
ul = 15,
ul_gprs = 15
)
)After running run_sim.R using the provided configuration, the output should look as shown here.
> source("~/github/MVNOCoreSim/run_sim.R")
INFO [2021-06-09 17:19:00] baseline -- 0 (nDev: 0, unattached: NA, time: 0)
INFO [2021-06-09 17:19:02] baseline -- 0.01 (nDev: 1874, unattached: 40, time: 0.1)
INFO [2021-06-09 17:19:03] baseline -- 0.02 (nDev: 3677, unattached: 48, time: 0.2)
INFO [2021-06-09 17:19:04] baseline -- 0.03 (nDev: 5434, unattached: 41, time: 0.3)
INFO [2021-06-09 17:19:05] baseline -- 0.04 (nDev: 7290, unattached: 38, time: 0.4)
INFO [2021-06-09 17:19:07] baseline -- 0.05 (nDev: 9090, unattached: 47, time: 0.5)
INFO [2021-06-09 17:19:08] baseline -- 0.06 (nDev: 10928, unattached: 51, time: 0.6)
INFO [2021-06-09 17:19:10] baseline -- 0.07 (nDev: 12726, unattached: 46, time: 0.7)
INFO [2021-06-09 17:19:12] baseline -- 0.08 (nDev: 14543, unattached: 46, time: 0.8)
INFO [2021-06-09 17:19:13] baseline -- 0.09 (nDev: 16366, unattached: 43, time: 0.9)
INFO [2021-06-09 17:19:15] baseline -- 0.1 (nDev: 18146, unattached: 35, time: 1)
...
INFO [2021-06-09 17:19:58] baseline -- 0.98 (nDev: 50000, unattached: 0, time: 2.79)
INFO [2021-06-09 17:19:58] baseline -- 0.99 (nDev: 50000, unattached: 0, time: 2.79)
INFO [2021-06-09 17:19:58] baseline -- 1 (nDev: 50000, unattached: 0, time: 2.79)
Here, baseline describes the name of the currently running simulation, which also dictated the folder in which the results are going to be stored. In this case, results are ending up in ./data/baseline/.
Next, the output shows the percentage progress towards completion based on the configured runtime. nDev shows the number of generated devices, unattached shows the number of generated devices that have not yet completed their attachment cycle. time provides the passed simulation time in minutes.
The simulation creates five output files in the ./data/baseline/ folder. Example results are included in this repository in csv.gz format. These example results have been generated using the preconfigured run_sim.R script.
arrivals.csv.gz contains a ledger of each signaling dialog and when it was processed by which component. The format is default to r-simmer and well documented there.
arrivals_global.csv.gz contains a ledger of all arrivals, including devices with their start and end time.
attributes.csv.gz contains the attributes of both devices and signaling messages as well as global attributes that keep track of the general simulation state. The structure is once again common to r-simmer.
resources.csv.gz contains a ledger of the available resources, their queue sizes as well as occupied CPUs during the simulation time.
config.csv contains a description of the input configuration for later referencing.
For questions or issues with the simulation, please raise an issue or contact Stefan Geissler (stefan.geissler@informatik.uni-wuerzburg.de) or any of the authors of the related publication.