Page 42 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

4.5 0.03
Reconfiguration - BE τ = 2
4 Reconfiguration - ST τ = 2 Reconfiguration - BE τ = 3
Reconfiguration - ST τ = 3 0.025 Reconfiguration - BE τ = 4
Reconfiguration - BE τ = 5
Reconfiguration - ST τ = 4
3.5 No Reconfiguration - ST τ = 2 0.02 No Reconfiguration - BE τ = 2
Reconfiguration - ST τ = 5
Throughput (Gbps) 2.5 No Reconfiguration - ST τ = 4 0.015
No Reconfiguration - BE τ = 3
Packet Loss Ratio
No Reconfiguration - BE τ = 4
3
No Reconfiguration - ST τ = 3
No Reconfiguration - BE τ = 5
No Reconfiguration - ST τ = 5
2
1.5
1 0.01
0.005
0.5
0 0
2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20
Stream Mean Rate π (Streams/Second) Stream Mean Rate π (Streams/Second)
(a) Mid BE Traf ic Load = 1 Gbps
Fig. 16 – Centralized Bidirectional Topology: ST Total average through‑
put measured at the sink as a results of TAS with centralized con igura‑ 0.05 Reconfiguration - BE τ = 2
tion (CNC) management. 0.045 Reconfiguration - BE τ = 3
Reconfiguration - BE τ = 4
Reconfiguration - BE τ = 5
0.04 No Reconfiguration - BE τ = 2
are in general more costly in terms of sent and received 0.035 No Reconfiguration - BE τ = 3
Packet Loss Ratio
No Reconfiguration - BE τ = 4
CDT frames in the network. Therefore, the higher the ad‑ 0.03 No Reconfiguration - BE τ = 5
mission rate, the more overhead is observed in the con‑ 0.025
trol plane, though based on Fig. 15, the overall overhead is 0.02
well below 1 Mbps and therefore is minuscule compared 0.015
to the channel capacity. We also observe from Fig. 15 that 0.01
the results for different stream lifetimes differ only very 0.005
slightly since for any value, almost all the streams are 0
12
6
10
8
14
16
accepted, generating the same total overhead. 2 4 Stream Mean Rate π (Streams/Second) 18 20
Fig. 16 shows the average overall throughput measured at
(b) High BE Traf ic Load = 2 Gbps
the ST sinks for the bidirectional ring topology. Compared
to the unidirectional ring (see Fig. 10), the throughput for Fig. 17 – Centralized Bidirectional Topology: BE frame loss ratio for TAS
the bidirectional ring is much higher, typically increased with centralized con iguration (CNC) management.
by a factor of two. tion parameters shown in Table.1.
Similar to the unidirectional ring topology, the bidirec‑
tional topology achieves zero loss for ST streams while 5.3.1 Unidirectional ring topology
signi icantly reducing the BE packet loss rate. Fig. 17
shows the BE packet loss ratio for the bidirectional ring The decentralized model essentially transfers some of
network. The maximum BE loss for the high BE traf ic in‑ the CNC functions (e.g., TAS recon iguration and resource
tensity = 2.0 is around 30% which is a signi icant re‑ reservation modules) from the centralized model down

duction from the unidirectional topology (of around 90%, to the TAS enabled egress ports of the TSN switches in the
see Fig. 11). data plane. The main difference between the centralized
In contrast to the unidirectional topology, the bidirec‑ and decentralized models is the signaling performance
tional topology with central (hybrid) CNC recon igura‑ which is now in‑band and can affect data traf ic. In addi‑
tion achieves improved QoS metrics and admission rates. tional evaluations that are not included due to space con‑
Overall, the ST traf ic throughput is typically doubled in straints, we have found that with the in‑band CDT traf ic
the bidirectional ring network compared to the unidirec‑ in the decentralized model, the average ST and BE packet
tional ring network. We can thus conclude that our pro‑ delays are about the same as the centralized model in
posed centralized (hybrid) CNC recon iguration can effec‑ Fig. 6. Typically, the ST stream’s average delay is mini‑
tively utilize the higher capacity provided by the bidirec‑ mal to near constant for both the recon iguration and “no
tional ring network for dynamic ST traf ic, with random recon iguration” approaches. For BE, the “no recon igura‑
ST low generations and random ST low lifetimes. tion” approach produces constant average delay for each
BE traf ic intensity.

5.3 Decentralized model evaluation Fig. 18 shows the maximum ST packet delay for the uni‑
directional ring network using the decentralized model.
Analogous to the centralized (hybrid) recon iguration In contrast to the average ST packet delay, the maximum
evaluation, we evaluate our proposed decentralized re‑ delay is affected by the in‑band CDT traf ic. In the decen‑
con iguration from Section 4 with both periodic ST traf‑ tralized model, the CDT traf ic is given the highest prior‑
ic and sporadic (random) BE traf ic, as speci ied in Sec‑ ity above both ST and BE traf ic. Therefore, the maximum
tion 5.1.2. As before, we evaluate the network with TAS delays can reach about 150 s, which is somewhat higher
shaper on the industrial control loop unidirectional and than for the centralized recon iguration in Fig. 7, but still
bidirectional topology and collect results for the simula‑ well below 1 ms.

26 © International Telecommunication Union, 2021