Source-linked AI summary
Sparse Distributed Fiber Optic Sensor Placement in Metropolitan and Datacenter Elastic Optical Networks
Sleman Mouammar, Ítalo B. Brasileiro, André Drummond
TL;DR
Sparse DFOS placement could reduce the number of sensing devices, but its effects on network performance and survivability remain a research gap. This paper evaluates placement heuristics and prediction windows in metropolitan and data center EONs. GSPC substantially reduces affected circuits at low coverage, while 25% coverage performs comparably to full coverage.
Problem
The performance and survivability effects of sparse DFOS placement remain a research gap, despite its potential to reduce the number of deployed sensing devices.
Source: §1 · pp. 1–2Method
Results
With 5% DFOS link coverage, GSPC reduces affected circuits by 81.3% in ION and 59.4% in DC; 25% coverage performs comparably to full coverage.
Source: §1 · p. 2, B. Results and Discussion · p. 6Takeaways & Limitations
Monitoring 25% of links can deliver performance comparable to full DFOS coverage, offering an opportunity for significant CapEx savings.
Source: §1 · p. 2Failure prediction is assumed to be 100% accurate, and false positives are outside the study's scope.
Source: §3 · p. 4
Abstract
from arXiv · showhide
Distributed Fiber Optical Sensing (DFOS) leverages optical backscattering signals to predict failure events and enable proactive Just-In-Time (JIT) restoration. Modern network infrastructures have optical components and fibers as key enablers of high throughput and low latency. The integration of sensing features in this optical infrastructure adds features such as network diagnosis, spatial positioning and sensing, and opens up opportunities for the implementation of intelligent network control expected for the next-generation networks. In this paper, we study how the quantity and the placement of the DFOS devices in an all-optical network can mitigate failure events. In our evaluation, we consider metropolitan and data center elastic optical networks. This work also evaluates sensing devices with various prediction time capabilities, in order to assess their effectiveness when considering a variable number of deployed devices and different placement heuristics. The evaluation uses key performance metrics, including the bandwidth blocking rate, number of affected circuits by failure events, and the service downtime. Simulation results demonstrate that the Greedy Shortest Path Coverage (GSPC) heuristic substantially reduces the number of circuits affected by failure when DFOS devices cover only 5% of links, while 25% coverage shows a performance gain comparable to the full coverage scenario, which is an opportunity for significant cost reduction. Finally, for the 5% sensing coverage scenarios, our results illustrate that prediction windows of 15 ms and 7.5 ms are sufficient for JIT restoration in metropolitan and data center networks, respectively.
1. INTRODUCTION
The paper investigates how sparse DFOS placement can support failure prediction and restoration while reducing the number of deployed sensors. It evaluates placement strategies and prediction windows in metropolitan and data center EONs.
1. INTRODUCTION
- ISAC combines environmental sensing and data communication in shared waveform, spectrum, and hardware, and DFOS provides network sensing capabilities.The introduction connects this paradigm to network diagnosis, positioning, and sensing, and cites its relevance to 6G services such as fast recovery and high availability.Source: §1 · p. 1
- Although sparse DFOS placement is NP-hard and depends on traffic, topology, and device efficiency, its effects on network performance and survivability remain a research gap.DFOS can monitor tens of kilometers of fiber, motivating sparse deployment to reduce the number of sensing devices.Source: §1 · pp. 1–2
- In an illustrative failure scenario, sensing at node D enables proactive reallocation of all three circuits affected on the central link, whereas sensing only at A reallocates one-third.Earlier work with DFOS at every node reported more than 90% fewer circuit disruptions in a metropolitan EON, but assumed full-node deployment.Source: §1 · p. 2
- 25% link coverage delivers performance comparable to full DFOS coverage, while 5% coverage with 15 ms prediction provides effective service maintenance.The authors identify significant CapEx savings from 25% coverage and report that 15 ms prediction is effective even with only 5% of links monitored.Source: §1 · p. 2
- The study evaluates sparse DFOS placement in metropolitan and data center EONs, adapts placement heuristics, proposes GSPC, and assesses devices with different failure-prediction windows.The evaluation examines how device quantity and placement affect failure prediction and mitigation.Source: §1 · p. 2
2. RELATED WORK
Related work addresses sensor placement, proactive failure prediction, restoration, and integrated sensing architectures. This paper builds on these strands to study DFOS placement for failure mitigation in EONs.
2. RELATED WORK
- Prior DFOS placement work reports a 26% improvement over a random-fit baseline with sensing lengths up to 80 km, while explore-and-pick reduces sensor count by 16% for full coverage.The cited studies consider multiple-link sensing and single-link failures with protection mechanisms, respectively.Source: §2 · p. 2
- Related approaches include ML-based WAN fault prediction with up to 88% accuracy, metropolitan sensing field trials, and telemetry for real-time consumption of sensing data.The paper assumes trained ML models for its experiments and notes that AI/ML methods could enhance its results.Source: §2 · p. 2
- Prior protection work targets recovery within 50 ms, whereas this study examines how varying DFOS quantity, placement, and reconfiguration time affect failure impact.The cited work proposes shared protection with time-constrained resource allocation in large-scale WDM networks.Source: §2 · p. 2
- The paper adapts ISAC sensing architectures for EONs and assumes sensing devices like the cited FSCD system, whose reported local decision latency is below 50 ns.The FSCD work combines SoP- and OTDR-based monitoring in sensing regions; cited responsiveness is under 1 µs.Source: §2 · p. 3
3. DFOS-ENABLED ARCHITECTURE IN EON
The proposed architecture integrates DFOS sensing into an EON fiber and coordinates detection and JIT restoration through an SDN-controlled management system. The study assumes accurate failure prediction and defines downtime using setup and switching times.
3. DFOS-ENABLED ARCHITECTURE IN EON
- The architecture places a DFOS device at a source node to monitor one unidirectional link, sensing every lightpath that traverses it while supporting simultaneous data transmission.The design is intended for metropolitan and data center optical networks and integrates the DFOS subsystem with EON components and an SDN control plane.Source: §3 · p. 3
- A dedicated in-band sensing frequency slot lets the sensing signal propagate alongside data lightpaths over the same fiber rather than requiring an out-of-band sensing fiber.The sensing slot is fixed at 12.5 GHz and separated from neighboring lightpaths by a guard band.Source: §3 · p. 3
- The DFOS head-end sends optical pulses into the monitored fiber and processes returning Rayleigh backscatter to obtain spatially resolved information about fiber conditions.The processing module extracts disturbances including vibrations, acoustic events, bending, and fiber cuts; the storage module retains measurements and timestamps for correlation with network events.Source: §3 · p. 3
- A failure-detection module passes DFOS disturbance features to an RMLSA module, which computes paths and spectral resources through the SDN control interface.The architecture places both modules under an SDN-controlled EON management system.Source: §3 · p. 4
- After detection, the controller identifies affected lightpaths and initiates JIT restoration; an unsensed link can still cause service interruption and delayed recovery.The study assumes failure prediction is 100% accurate and excludes false positives.Source: §3 · p. 4
- Circuit downtime combines cumulative setup time across allocation attempts with switching time; setup time depends on primary-path hops, propagation delay, and node-processing delay.For reactive restoration, switching time also includes failure-detection time and signaling along the broken primary path up to the failed link.Source: §3 · p. 4
4. DFOS DEVICE PLACEMENT HEURISTICS
The section compares five heuristics for sparse DFOS placement and presents GSPC, which greedily selects links to cover the largest number of unsensed shortest paths. Its selection phase runs for at most N device placements, with a stated overall time complexity based on the shortest-path set and link-path incidences.
4. DFOS DEVICE PLACEMENT HEURISTICS
- The heuristic comparison includes Most Used, Most Simultaneously Used, Node Degree First, Random Links, and the proposed GSPC.MU and MSU use preliminary traffic simulations; NDF uses node degree, while Random Links selects links randomly.Source: §4 · p. 4
- GSPC uses K-shortest-path routing to prioritize links shared by multiple shortest paths, increasing sensing coverage and the chances of reallocating more circuits.The method builds a link-to-path mapping and tracks which shortest paths remain unsensed as links are selected.Source: §4 · p. 4
- GSPC selects links with the highest coverage, defined as the number of currently unsensed shortest paths that traverse each link.It repeats this choice until all shortest paths are sensed or the available DFOS devices are used.Source: §4 · pp. 4–5
- The GSPC selection phase has time complexity O(NL), and the stated total complexity is O(V2(E + VlogV) + NL).Here, L is the total number of path-link incidences and N is the maximum number of DFOS-enabled links.Source: §4 · p. 5
A. Simulation Setup
The simulations model service requests, failure prediction, failure occurrence, and recovery in metropolitan and data center elastic optical networks. They use three topologies and evaluate blocking, failure disruption, prediction time, and downtime.
A. Simulation Setup
- Each simulation models 100,000 circuit requests over one year, with Poisson arrivals, exponentially distributed holding times averaging 600 s, and bandwidth demands from 10 to 400 Gbps.Results average 10 runs with different seeds and include 95% confidence intervals; failures comprise 100 uniformly timed link failures with a 4-hour mean time to repair.Source: A. Simulation Setup · p. 5
- The study uses ION and Catalunya metro topologies and a 64-node NovaCube data center topology; Fig. 3 depicts the studied topologies and link lengths.The average link lengths are 40.71 km for ION, 13.27 km for Catalunya, and 2 km for NovaCube.Source: Figure 3 · p. 5, A. Simulation Setup · p. 6
- Performance measures are bandwidth blocking rate, circuits affected by failures, prediction time, and downtime.The setup assumes a 6-minute failure-prediction window, limits reallocation attempts to 20 ms, and restricts sensing to links up to 100 km.Source: A. Simulation Setup · pp. 5–6
B. Results and Discussion
With DFOS on only 5% of links, GSPC reduces affected circuits more than random placement in both ION and data-center networks. Increasing coverage or prediction time further reduces impacts, with diminishing gains beyond roughly 25% coverage and topology-specific prediction thresholds.
B. Results and Discussion
- At the selected loads, restoration heuristics keep BBR below 5%, while traditional and full-coverage protection average 50% in ION and 20% in DC.Reserving one fixed spectrum slot per link for DFOS does not impact BBR.Source: Figure 4 · p. 6, B. Results and Discussion · p. 6
- 81.3% (ION) and 59.4% (DC): at 5% link coverage, GSPC reduces affected circuits versus traditional restoration.At roughly 25% coverage, reductions are 95.1% (ION) and 91.2% (DC); affected circuits per failure are 0.37 and 1.47, versus full-coverage values of 0.19 and 0.11, respectively.Source: B. Results and Discussion · p. 6, Figure 5 · p. 7, Figure 6 · p. 7
- Affected circuits decline as prediction time increases, then stabilize after roughly 15 ms in ION and 7.5 ms in DC.At those thresholds, GSPC yields 1.78 affected circuits per failure in ION and 5.9 in DC; further prediction time brings negligible improvement.Source: B. Results and Discussion · p. 6, Figure 5 · p. 7, Figure 6 · p. 7
- At 15 ms, GSPC reduces ION downtime by 20.3 ms (78.6%); at 5 ms, it reduces DC downtime by 18.2 ms (59.8%).Beyond those prediction times, additional time yields negligible benefits.Source: Figure 5 · p. 7, Figure 6 · p. 7, B. Results and Discussion · p. 7
- Across the metropolitan topologies, results show that distributing DFOS and selecting placement strategically matter; even random placement at 5% coverage mitigates failure impacts.GSPC provides greater benefits by selecting links covering more relevant shortest paths.Source: B. Results and Discussion · p. 7
6. CONCLUSION
The study evaluates DFOS placement strategies in metropolitan and data center EONs, focusing on sparse deployment and its effects on network performance and restoration. With GSPC, 5% link coverage substantially reduces affected circuits, while 25% coverage performs comparably to full coverage.
6. CONCLUSION
- The evaluation examines placement strategies in metropolitan and data center EONs using bandwidth blocking probability, circuits affected by failures, and service downtime.The study focuses on 5% of unidirectional links equipped with DFOS, the minimum deployment considered.Source: §6 · p. 7
- With GSPC, 5% link coverage substantially reduces affected circuits, while 25% coverage yields results comparable to full network coverage.These results concern the number of circuits affected by failure events.Source: §6 · p. 8
- For 5% deployment, prediction windows of 15 ms in metropolitan networks and 7.5 ms in data center networks are sufficient for JIT restoration.The paper identifies multi-failure scenarios, false alarms, missed detections, and pairing placement heuristics with routing solutions as future research directions.Source: §6 · p. 8