Factores moral y familiar

For the N:1 Protection with Physical Layer experiment [117], we implemented end-to-end software defined networking (SDN) management of the access and core network elements of a time-division multiplexing (TDM) dense wavelength division multiplexing (DWDM) long- reach passive optical network (PON). The physical layer in Figure 24 demonstrated co- existing heterogeneous services and modulation formats such as residential 10G PON channels, business 100G dedicated channel and wireless front haul on the same long reach TDM-DWDM PON system. This worked with both erbium doped fibre amplifiers (EDFAs) or semiconductor optical amplifiers (SOAs) for a TDM-DWDM PON up to 100km reach, 512 users and emulated system load of 40 channels.

Figure 23 - Switchover time (milliseconds) for 50 iterations of N:1 protection experiment

73

Figure 24 - Network level view of the demonstration

The Use case shown in Figure 25 exemplifies how path integrity in the Core and TDM- DWDM LR-PON based Access Metro network of a Telecommunications network could be assured through logical protection. The protection experiment demonstrated a dual-homed LR-PON protection mechanism where backup OLTs are shared among PONs in an N:1 scheme [107] and the service restoration is provided over an end-to-end SDN. The system carried out an initial phase of path-precomputation, where it sets up a backup path associated to the failure of a specific PON. The pre-calculation considers the input and output ports at the optical switch, the flow table configuration of the OF SDN switch (both access and core) and the configuration of the OLT flow table. The Failure Event (1) was caused by the feeder fibre between the primary OLT and first stage splitter on the PON being cut. This stops all upstream and downstream data on the Primary link. A hardware unit in the primary OLT FPGA monitors the upstream data path.

Figure 25 - Protection Experiment

5.3.2 Results

The first test of the protection experiment was the failure event emulated by using the optical switch to simulate a fibre cut in the backhaul fibre link between the primary OLT and the Access Network. Silence in the upstream activated a countdown timer in the primary OLT, which on expiry generated a failure detection and an in-band alarm to the controller of node 1. The duration of this timer took into account all normal silences on the PON due to the

N:1 Protection Scheme with PON Physical layer

74

1.25 milliseconds quiet windows and 1.25 milliseconds roundtrip time over maximum distance supported by the protocol of 125km, for a total of 2.5 milliseconds. The node 1 controller then alerted the overarching Network Orchestrator which calculated a path to restore services to the ONUs according to its knowledge of the full end-to-end topology covering the core and access networks. The Network Orchestrators were then instructed by the Network Orchestrator to provision the protection path through the backup OLT. Figure 26 shows a capture of the message flow for one of the protection experiment runs.

Figure 26 - Protection Message Flow

Figure 27 shows the service restoration time for the protection mechanism conducted where backup OLTs are shared among PONs in an N:1 scheme [117]. The baseline time between the two paths of approximately 50 milliseconds is given when the switchover is proactively triggered by the controller, without waiting for a failure event. In contrast, the protection results show the restoration time when a failure event is caused by a cut in the backhaul link between the primary OLT and the Access Node. Silence in the upstream activates a countdown timer in the primary OLT controller, which on expiry generates a failure detection and an in-band alarm to the Openflow access Network Controller. The access Network Controller alerts the Network Orchestrator, which provisions the protection path. The average protection time is measured at 64 milliseconds, with variations between 50 and 100 milliseconds attributed to the random delay in the failure detection.

75

Figure 27 - Protection Timings

The N:1 + physical layer experiment was executed secondly [118], where the detection response was optimised. The results of the service restoration time for the SDN control plane based protection mechanism are shown in Figure 28. The average restoration time over 70 measurements was 41 milliseconds.

Figure 28 - Service restoration time for the protection mechanism and the DWA through the implemented SDN control plane

In Figure 29, we show the breakdown of the various timings that comprise the 41 milliseconds protection figure. The hardware monitoring at the OLT can detect a failure in the network in about 2.5 milliseconds. A further 1 milliseconds is taken for the alarm packet

Summary

76

to be created and sent to the Core Node switch. The time needed by the protocol to re- establish downstream synchronization is between 2 and 3 milliseconds.

Figure 29 - Timings Trace

We know that some time may be needed to re-range the ONUs in addition to the synchronization time (between 2 and 4 milliseconds), however in this work we assume that ranging to the backup OLT can be done during normal operation of the PON [116]. Intra- control plane communication is done through a dedicated network with typical latencies. The network latencies between both the OLT and the Network Orchestrator and the Network Orchestrator and the Network Controllers are emulated in the test-bed and set at 4 milliseconds each. The latency and the processing times for both the Network Controllers is also emulated as 5 milliseconds each. The core network recovery happens in parallel to the access network recovery time.

Accordingly, within 15 milliseconds of the failure, the optical and electronic switch components and the backup OLT have been instructed to reconfigure their protection paths. Within 33 milliseconds after the failure, the electronic switch components within the core and access are configured, and by 38 milliseconds, the optical switch component is configured. In order to understand the effect of centralising both the Network Orchestrator and the Network Controllers, we compared the above results with the case where orchestrator and controllers are collocated within the Core Network. This was accomplished by setting the emulated intra-control plane latencies at zero. The results are shown in Figure 28 as the basic protection line. On average, basic protection can be accomplished within 27.8 milliseconds.