Page 37 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1

a
l
i
r
C Centralized Network k Table 1 – Simulation parameters
en
t
z
w
o
r
et
ed

N
u
g
a
r
i
o
C Configuration (CNC) )
f
n
C
(
C
N

i
t
n
o
Traffic Traffic Key Symbol Value
Sink Source
Simulation Dura‑ limit 100 seconds
Full Duplex tion
TSN
Switch Ethernet Initialized Cycle 50 s
Time
Traffic Link Traffic Initialized Gating 20% (i.e., 10 s)
Source Source Ratio init
TSN TSN
Traffic Switch Switch Traffic Average Streams 1 − 20
Sink Sink per Second
Average stream du‑ 2 − 5 seconds
ration
Traffic Traffic BE Traf ic Intensity 0.1, 1.0, 2.0 Gbps
Source TSN TSN Source (580 byte packets)
Switch Switch ST sources 6
Traffic Traffic Queue Size 512 KB
Sink Sink
TSN TSN 5. PERFORMANCE EVALUATION
Switch Domain
Traffic Traffic 5.1 System overview and simulation setup
Sink Source
Fig. 5 – Industrial control loop topology [36]: Each source generates This section explains the simulation setup and model.
stream data with varying hop counts and packet rates unidirectionally Furthermore, the topology and simulation scenarios will
or bidirectionally across the six switches ultimately destined to a sink
be presented. Throughout, we employ the OMNet++ [86]
10% and 90% for the lower and upper limits, respectively. simulation environment. For each evaluation for a given
The main reasoning behind this design choice is to avoid set of parameters, we conduct 5 independent simulation
any starvation of lower priority traf ic. replications; each replication simulates the network for
20 seconds. The widths of the resulting 95% con idence
intervals are smaller than 5% of the corresponding sam‑
4.1.6 Path computation ple means and are therefore omitted from the plots to
avoid clutter.
While a path computation module is fundamentally nec‑
essary in any switch (in a decentralized/distributed net‑ 5.1.1 Network model
work), we de ine static shortest path routing tables
for destination addresses and associated ports on each The network topology is modeled around an industrial
switch. Essentially, we assume a procedure to compute control loop topology that consists of six core switches in
paths, i.e., we assume that there is a path computation a ring topology. In the case of the centralized model, a
module, e.g., Path Computation Engine (PCE), that is used CNC is used with out‑of‑band connections to each of the
in both centralized and distributed con iguration models core switches; while in the distributed approach, the sig‑
(the path computation can be accelerated with hardware naling is in‑band and can interfere with data traf ic within
modules [47,70,83], if needed). We make this assumption the TSN domain, as shown in Fig. 5. Each switch‑to‑switch
to simplify operations and place emphasis on the TAS re‑ link operates as a full‑duplex Ethernet link with a capac‑
con iguration technique. ity(transmissionbitrate) = 1Gbps. Eachswitchcanact
as a gateway for a number of traf ic sources and one sink.
The distance between two successive switches along the
4.1.7 Network resource table ring is ixed to 100 m and the switch‑to‑switch propaga‑
tion delay is set accordingly to 0.5 s. The out‑of‑band
To remove certain overheads of the con iguration pro‑ connections have exactly the same con igurations as the
cedure, the network resource table operates in tandem normal full‑duplex Ethernet links in the data plane, i.e.,
with the stream registration table to accurately determine the same bitrate and propagation delay. All switches are
the required network resources (mainly bandwidth for con igured to use 802.1Qbv TAS as the traf ic shaper for
our traf ic model) per switch egress port. The network each switch‑to‑switch egress port whose low schedule
resource table classi ies streams based on periodic and (STgatingratioandcycletime)is con iguredbytheCNCin
sporadic stream properties, though currently our focus the centralized (hybrid) model and independently in the
is on periodic ST streams. Any stream that has been ap‑ decentralized (fully distributed) model. For all simulation
proved by a switch has an associated record in the net‑ runs, the ST slot size is initialized to 20% of the CT. For
work resource table, located within each switch, which the operation without recon iguration, the ST slot size is
can be called to compute and store current and remain‑ kept at 20% of the CT; whereas, for the operation with re‑
ing link/port loads for each switch. Each egress port has con iguration, the ST slot size is dynamically recomputed
a network resource table. when the irst stream transmission request arrives.
© International Telecommunication Union, 2021 21