Page 40 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

0.8 operation, the ST maximum packet delays hover around
0.7 55–101 s, see Fig. 7). This stronger increase of the max‑
imum ST packet delays is a result of the BE packet traf‑
0.6 ic interfering with the ST packet traf ic due to the lack
Packet Loss Ratio 0.4 of TAS operation. In particular, ST packets are blocked
0.5
from transmission during an ongoing transmission of a
580 byte BE packet (as we considered non‑preemptive
0.3
Reconfiguration - BE τ = 3
Reconfiguration - BE τ = 4
0.2 Reconfiguration - BE τ = 2 priority scheduling). Second, since no admission control
Reconfiguration - BE τ = 5
No Reconfiguration - BE τ = 2 based on TSN slot reservation is used, congestion arises
0.1 No Reconfiguration - BE τ = 3
No Reconfiguration - BE τ = 4
No Reconfiguration - BE τ = 5 for ST traf ic loads of = 6 to 20 ST streams per second,
0
2 4 6 8 10 12 14 16 18 20 causing high mean and maximum delays for both ST and
Stream Mean Rate π (Streams/Second) BE traf ic. Third, due to the congestion, packet drops oc‑
(a) Mid BE Traf ic Load = 1 Gbps cur at high ST loads for both ST and BE packet traf ic. We
0.9 also note that since no signaling traf ic is used, the priority
0.8 scheduling benchmark without TSN operation provides a
performance reference for both the centralized and the
0.7 Reconfiguration - BE τ = 2 decentralized TSN model.
Packet Loss Ratio 0.5 No Reconfiguration - BE τ = 2 Overall, we conclude that the proposed centralized (hy‑
0.6
Reconfiguration - BE τ = 3
Reconfiguration - BE τ = 4
Reconfiguration - BE τ = 5
brid) recon iguration approach provides a means to en‑
No Reconfiguration - BE τ = 3
sure that dynamically varying numbers of ST streams are
0.4
No Reconfiguration - BE τ = 4
0.3 No Reconfiguration - BE τ = 5 accommodated as permitted by the available link capacity
in the unidirectional ring network. However, the unidi‑
0.2
rectional ring network does not involve any distinct rout‑
0.1
2 4 6 8 10 12 14 16 18 20 ing choices towards the destination. In order to examine
Stream Mean Rate π (Streams/Second) the performance of the proposed centralized recon igu‑
(b) High BE Traf ic Load = 2 Gbps ration in a network with different routing paths, we next
consider the operation of the ring network topology as a
Fig. 11 – Centralized Unidirectional Topology: BE frame loss ratio for bidirectional ring network.
TAS with centralized con iguration (CNC) management.
creases. For the “no recon iguration” approach, the BE 5.2.2 Bidirectional ring topology
packet loss is typically constant even for high loads of BE
traf ic. The unidirectional ring topology certainly simpli ies the
calculation of the ST slot window in the recon iguration.
For a benchmark comparison of the TSN effectiveness, In order to examine whether the proposed centralized
and speci ically TAS, we conducted additional evaluations (hybrid) recon iguration approach can ef iciently utilize
for the scenario in Fig. 6 without the TSN slot reservation, the higher capacity of a more complex network with mul‑
admission control, and TAS scheduling. Speci ically, we tiple routing options, we examine the bidirectional ring
considered an ST stream mean generation rate of 1–20 network. In the bidirectional ring network, each two‑port
streams per second with a mean lifetime = 5 seconds switch has now two paths to the destination. We em‑
with the mid and high BE traf ic loads of = 1.0 Gbps ploy shortest path routing according to the hop count. We
and 2.0 Gbps. We employed strict priority scheduling at set the edge link (source to irst ring switch and last ring
each switch without any TSN slot reservation, i.e., each switch to sink) capacities to 2 Gbps to avoid congestion on
switch output port schedules and transmits all ST packets the edge links (which the CNC does not control).
before any BE packets. We outline three main observa‑ Fig. 12 shows the average mean ST and BE packet delay
tions for the unidirectional ring topology. First, while the for different stream lifetimes . Compared to the unidi‑
mean delays were generally very low for ST traf ic (34– rectional topology (see Fig. 6), the bidirectional signi i‑
55 s for the low traf ic load range = 1 to 5 ST streams cantly reduces the packet delay since an extra port with
per second), the priority scheduling of the ST packets can full‑duplex link support now provides extra capacity to
severely starve the low‑priority BE traf ic (for the high service streams giving more slot reservations to BE even
= 2.0 Gbps BE load, the mean BE packet delays in‑ at high ST stream loads.
crease from a minimum of 15 ms to a maximum of around Fig. 13 shows the maximum ST packet delays for the bidi‑
0.1 s as the ST load increases from 1 to 20 streams per rectional ring topology with CNC. We observe from Fig. 13
second; whereas, with TSN, the mean BE packet delays in‑ in comparison with the corresponding maximum packet
crease from around 10 ms to 21 ms, which is outside the delayplotfortheunidirectionalringinFig.7, thatthebidi‑
plotted range of Fig. 6(b)). Additionally, compared to TSN, rectional topology with con iguration gives higher maxi‑
the maximum delays and jitter increase more strongly as mum packet delays, which is mainly due to the substan‑
the BE and ST loads increase (the ST maximum packet tially increasing ST stream acceptance, as examined next
delays range from 34 s to 20 ms; while, with the TSN in Fig 14. The “no recon iguration” keeps the ST slot size

24 © International Telecommunication Union, 2021