Page 43 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

0.18 20
Reconfiguration - ST τ = 2
0.16 18 Reconfiguration - ST τ = 3
Reconfiguration - ST τ = 4
Signaling Overhead (Mbps)
16 No Reconfiguration - ST τ = 2
Reconfiguration - ST τ = 5
Max Packet Delay (ms) 0.12 No Reconfiguration - ST-Max τ = 2 12 No Reconfiguration - ST τ = 5
0.14
No Reconfiguration - ST τ = 3
14
No Reconfiguration - ST τ = 4
Reconfiguration - ST-Max τ = 2
Reconfiguration - ST-Max τ = 3
Reconfiguration - ST-Max τ = 4
Reconfiguration - ST-Max τ = 5
0.1
10
No Reconfiguration - ST-Max τ = 3
8
0.08
No Reconfiguration - ST-Max τ = 4
No Reconfiguration - ST-Max τ = 5
0.06
4
0.04 6
2
0.02 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)
Fig. 18 – Decentralized Unidirectional Topology: Max delay for TAS. Fig. 20 – Decentralized Unidirectional Topology: Stream Signaling Over‑
head for TAS.
14
Reconfiguration - ST τ = 2 ized model in Fig. 20 with the centralized model in Fig. 9
Reconfiguration - ST τ = 3
Reconfiguration - ST τ = 4
12 No Reconfiguration - ST τ = 2 indicates that the decentralization increases the signaling
Mean Signaling Delay (υs) 10 No Reconfiguration - ST τ = 4 aggregate signaling overhead bitrate in the decentralized
Reconfiguration - ST τ = 5
overhead by over two orders of magnitude. However, the
No Reconfiguration - ST τ = 3
No Reconfiguration - ST τ = 5
model is still below 20 Mbps and thus below 2% of the
8
1 Gbps link capacity.
6
pared to the unidirectional centralized model (cf. Fig. 10)
4 Throughput results are generally the same when com‑
and are therefore omitted. Similarly, the packet loss rate
2 4 6 8 10 12 14 16 18 20 is nearly similar to the unidirectional centralized model
Stream Mean Rate π (Streams/Second) (cf. Fig. 11). However, the unidirectional topology with ei‑
ther the centralized or decentralized approach generally
Fig. 19 – Decentralized Unidirectional Topology: Average stream signal‑ gets bottlenecked at lower traf ic loads compared to the
ing delay for TAS.
bidirectionalringnetwork. Therefore, BEtraf icsuffersas
The stream admission rate for the decentralized model is more ST streams request TAS slot reservations. We next
very similar to the centralized model (see Fig. 8) and is examine the bidirectional ring network for decentralized
not displayed in detail due to space constraints. Fig. 19 operation to determine how the BE traf ic performance
shows the signaling delay for ST stream registration in the can be improved while maintaining the ST traf ic perfor‑
decentralized model. In contrast to the centralized model, mance.
the decentralized model’s in‑band CDT traf ic implies var‑
ied stream signaling delays. As the streams generation 5.3.2 Bidirectional ring topology
rate increases, the overall average signaling delay de‑
creases which is due to the increased rejections as more For the bidirectional topology using the decentralized
streams attempt to request network resources. In the de‑ model we found that the in‑band CDT traf ic affects the
centralized model, a rejection by an intermediate bottle‑ data traf ic similar to the decentralized unidirectional
necked switch implies a termination of the reservation at‑ model, i.e., maximum ST packet delay is somewhat in‑
tempt and a noti ication to any previous pending stream creased while the mean ST packet delay is essentially un‑
records to cancel the potential reservation and eventually changed. As the ST stream lifetime is increased, i.e., the
notify the source of the rejection. If this rejection happens number of ST streams at any time increases, the BE slots
closer to the source, then the average signaling delay will are reallocated to ST streams which increases the mean
be shorter compared to a stream acceptance. In general, BE packet delay which is similar to the centralized model
the average stream signaling delay is on the order of mi‑ (cf. Fig. 12) and is therefore omitted.
croseconds which is reasonable for most industrial con‑ Fig. 21 shows the maximum ST packet delay. While the re‑
trol systems applications. con iguration approach looks very similar to the central‑
Generally, the decentralized model produced greater sig‑ ized model (cf. Fig. 13), the no recon iguration approachis
naling overhead than the centralized model (cf. Fig. 9) affected by the in‑band CDT traf ic which raises the maxi‑
since CDT traf ic is measured at each data switch traf ic mum ST packet delay in some no recon iguration scenar‑
port for incoming and outgoing as shown in Fig. 20. Anal‑ ios to around 100 s.
ogous to the signaling delay, the more ST streams are ac‑ The admission rate is exactly the same as for the central‑
cepted, the more overhead is observed. Therefore, as the ized model (cf. Fig. 14). Fig. 22 shows the average signal‑
stream lifetime increases and consequently, the more ing delay for ST stream registration. Similar to the unidi‑
rejections occur, the lower the overhead. Overall, the rectional topology, the mean signaling delay starts to de‑
comparison of the signaling overhead for the decentral‑ crease as the load increases due to higher rejections.

© International Telecommunication Union, 2021 27