Page 38 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

5.1.2 Traf ic model 0.45
We consider periodic (pre‑planned) traf ic for the ST traf‑ 0.4 No Reconfiguration - BE-0.1ρ L
Mean Packet Delay (ms)
ic and sporadic (random) Poisson traf ic for the BE traf‑ 0.35 No Reconfiguration - BE-1.0ρ L
No Reconfiguration - BE-2.0ρ L
Reconfiguration - BE-0.1ρ L
ic. To emulate dynamic conditions in the network, we 0.3 Reconfiguration - BE-1.0ρ L
Reconfiguration - BE-2.0ρ L
employ several distributed ST sources that generate ST 0.25 No Reconfiguration - ST-0.1ρ L
No Reconfiguration - ST-1.0ρ L
streams according to the network and traf ic parameters 0.2 No Reconfiguration - ST-2.0ρ L
Reconfiguration - ST-0.1ρ L
shown in Table 1. The stream generation follows a Pois‑ 0.15 Reconfiguration - ST-1.0ρ L
Reconfiguration - ST-2.0ρ L
son process with a prescribed mean rate of generated 0.1
streams per second. We refer to the stream generation 0.05
rate also as the stream mean rate . Each ST stream in‑ 0
jects one packet of size 64 bytes per cycle. A destination 2 4 6 8 10 12 14 16 18 20
address is assigned by the number of switch‑to‑switch Stream Mean Rate π (Streams/Second)
hops. A given stream that has been generated at the traf‑ (a) = 2
ic source attached to a given TSN switch is destined to the 0.45
traf ic sinks at the other ive TSN switches with a uniform 0.4 No Reconfiguration - BE-0.1ρ L
probability of 1/5. Furthermore, at stream creation, each 0.35 No Reconfiguration - BE-1.0ρ L
No Reconfiguration - BE-2.0ρ L
stream is given a start time (usually the current runtime), 0.3 Reconfiguration - BE-0.1ρ L
Reconfiguration - BE-1.0ρ L
Reconfiguration - BE-2.0ρ L
and a inish time according to a stream lifetime (dura‑ 0.25 No Reconfiguration - ST-0.1ρ L
No Reconfiguration - ST-1.0ρ L
tion) that follows the exponential distribution with mean Mean Packet Delay (ms) 0.2 No Reconfiguration - ST-2.0ρ L
Reconfiguration - ST-0.1ρ L
. The exponential stream lifetime is considered as call Reconfiguration - ST-1.0ρ L
0.15 Reconfiguration - ST-2.0ρ L
level dynamics in communication networks often follow 0.1
Poisson process dynamics, i.e., exponential call lifetimes.
As TSN networks become more commonly deployed, it 0.05
will be important to verify the stream lifetime dynamics 0 2 4 6 8 10 12 14 16 18 20
through real system measurements. Stream Mean Rate π (Streams/Second)
We consider admission as the completion of the reserva‑ (b) = 5
tion of the network resources for the low from the source
node to the destination node. Each source is attached to Fig. 6 – Centralized (Hybrid) Unidirectional Topology: Mean end‑to‑end
delay for ST and BE traf ic for mean stream durations = 2 and =
a core TSN switch gateway ( irst hop switch). While the
5 seconds under different BE loads , and ST stream rates .
TSN switches operate with time synchronization, the ST
sources (outside the TSN domain) do not need to be syn‑ for both BE and ST traf ic. Since the CNC manages the ST
chronized. However, note that the ST traf ic follows an traf ic streams and therefore guarantees the bandwidth
isochronous traf ic class, as speci ied by IEC/IEEE 60802, rates needed across the stream’s path, the ST delays are
whereby the sources are synchronized with the network less than 100 s for all average stream durations . The
after stream registration is completed. In particular, the ST streams with recon iguration at the CNC experience
ST sources inject the ST traf ic in just‑in‑time fashion, i.e., higher delays than for the no recon iguration scenarios
the transmission of the ST packets out of a source starts since we essentially push more ST traf ic into the network
at the instant of the start of the ST transmission slot at which increases the queuing delay. BE traf ic experiences
the switch that is directly attached to the source. Pack‑ much higher delays than ST. With the no recon iguration
ets are time stamped for the packet delay measurement approach, the BE traf ic delay is nearly constant since the
at the time instant when the packet transmission out of gating ratio is left unchanged. The BE mean delay in‑
the source commences. creases dramatically (up to 21 ms) with recon iguration
since the accepted ST streams tend to consume the full
5.2 Centralized (hybrid) model evaluation permitted 90% of the CT, leaving only very limited trans‑
mission resources for BE traf ic.
In evaluating the proposed solution described in Sec‑
tion 3, we consider both periodic ST traf ic and sporadic As mentioned in the introduction section, TSN needs to
BE traf ic, as described in Section 5.1.2. We evaluated limit the maximum delay in order to deterministically for‑
the CNC with TAS shaper on the industrial control loop ward traf ic across a TSN domain. Fig. 7 shows the maxi‑
for the unidirectional and bidirectional topologies and re‑ mum delay for the ST traf ic. We observe from Fig. 7 that
sults are collected for the simulation parameters shown in for the unidirectional ring topology with a maximum of
Table 1. ive hop streams, the maximum delay for the recon ig‑
uration approach is below 0.11 ms. On the other hand,
5.2.1 Unidirectional ring topology for the “no recon iguration” approach, the maximum de‑
lay is below 60 s due to lower frame residence time on
Fig. 6 shows the average mean delay for ST traf ic and for each switch; however, recon iguration increases the ad‑
BE traf ic for the centralized unidirectional ring topolo‑ mission of ST streams as examined next in Fig 8. TAS
gies. The average delays are generally short and stable in conjunction with the CNC registration and reservation

22 © International Telecommunication Union, 2021