Now 5061 visitors
Today:0 Yesterday:0
Total: 147058 212S 50P 4R
2026-10-20, Week 43
Member Login
TACT Journal Homepage
Call for Paper
Committee
Find My Paper
Author Homepage
Statistics & History
Welcome Message
Paper Procedure
FAQ
Paper Archives
Outstanding Papers
Program & Proceedings
Presentation Platform
Hotel & Travel Info
Photo Gallery
Scheduler Login
Seminar
Archives Login
Sponsors


















Presentation Assistant Go to [Q&A Session]
Paper Number
Paper Title
Keyword
ICACT20260346 Slide.01        [Big Slide]      [YouTube] Chrome Click!!
Hello everyone. Today I will present our work, “UW-LFRR: Loop-Free and Localized Fast Reroute in Underwater Ad Hoc Networks.” We focus on a very specific but critical question: when an underwater network suffers sudden multi-node failures and the routing control plane has not converged yet, how can the data plane keep forwarding immediately? The core idea of UW-LFRR is straightforward: instead of waiting for control-plane updates, each node leverages only its available local information to perform fast repair directly in the data plane.
ICACT20260346 Slide.02        [Big Slide]      [YouTube] Chrome Click!!
Underwater ad hoc networks are widely used for environmental monitoring, AUV cooperation, and mission-critical underwater operations. However, acoustic communication has fundamental physical constraints: slow propagation, very limited bandwidth, high bit-error rates, and strict energy budgets. As a direct consequence, control messages disseminate slowly and broadcasting is expensive, so after failures happen, the control plane often cannot converge in time in underwater environments.
ICACT20260346 Slide.03        [Big Slide]      [YouTube] Chrome Click!!
A useful way to understand the challenge is to contrast underwater acoustic networks with terrestrial radio networks. In radio networks, signals propagate quickly and control information can be refreshed promptly, so relying on the control plane after a failure is often reasonable. In underwater networks, propagation is slow and links are fragile; spreading control information across the network comes with a high cost. This is why, once failures occur, we cannot assume the control plane will repair routes immediately, and we need an instant data-plane response.
ICACT20260346 Slide.04        [Big Slide]      [YouTube] Chrome Click!!
This motivates the concept of the Convergence Blind Zone, or CBZ. CBZ is the vulnerable period where the physical topology has already changed, but routing control information has not been updated yet. During CBZ, packets continue to follow pre-failure routes, repeatedly pointing to failed next hops, which creates blackholes, queue buildup, and severe energy waste. The real problem is not that no path exists, but that a new path cannot be computed and installed fast enough before convergence completes.
ICACT20260346 Slide.05        [Big Slide]      [YouTube] Chrome Click!!
CBZ therefore leads to significant packet loss and energy waste, and in underwater networks the convergence time is often long. Under tight energy constraints, a severe CBZ event can even push nodes into permanent downtime. For this reason, our goal is not to chase a globally optimal route after convergence, but to keep the network usable throughout CBZ.
ICACT20260346 Slide.06        [Big Slide]      [YouTube] Chrome Click!!
In underwater environments, failures are often correlated in space or time, for example when multiple nodes deplete their batteries around the same time, equipment malfunctions occur in bursts, or localized physical damage takes out several nodes at once. Sudden multi-node failures can rapidly enlarge blackhole regions, fragment the topology, and isolate otherwise healthy nodes, creating cascading effects. That is why a recovery mechanism must be able to stop the bleeding quickly, rather than waiting for network-wide reconvergence.
ICACT20260346 Slide.07        [Big Slide]      [YouTube] Chrome Click!!
Existing approaches face clear limitations underwater. Proactive routing schemes suffer from long convergence delays due to long RTTs, reactive schemes rely heavily on broadcast and can trigger severe overhead, and terrestrial fast reroute designs are often mismatched: precomputed backups become fragile under multi-node failures, while failure-carrying or similar approaches can impose heavy per-packet computation that is unsuitable for energy- and computation-constrained underwater nodes. What underwater networks need is a lightweight, localized, data-plane-driven recovery mechanism.
ICACT20260346 Slide.08        [Big Slide]      [YouTube] Chrome Click!!
UW-LFRR is designed with four goals in mind: fast recovery, localized computation, loop-free forwarding, and minimal control overhead. The central philosophy is to push recovery down into the data plane, use a pre-failure topology snapshot together with local failure perception, and solve only the truly affected local problem instead of triggering network-wide actions.
ICACT20260346 Slide.09        [Big Slide]      [YouTube] Chrome Click!!
At a high level, UW-LFRR proceeds in three steps. It first constructs a purified local view by removing nodes that are known to be unreachable. It then applies a two-stage relay selection strategy: it tries to quickly graft traffic back onto the original shortest-path “backbone,” and only if that is impossible does it perform a constrained fallback search inside the affected region. Finally, it uses a first-packet paving mechanism so that temporary forwarding state can be installed rapidly in the data plane, without requiring control-plane participation.
ICACT20260346 Slide.10        [Big Slide]      [YouTube] Chrome Click!!
Before any failures, each node maintains a network-wide topology snapshot and builds a local shortest path tree based on hop count. Once failures occur, the snapshot is inevitably stale, but UW-LFRR does not attempt to enforce strong consistency during CBZ. Instead, it explicitly accepts that inconsistency exists and applies conservative local corrections based on the failures the node can perceive.
ICACT20260346 Slide.11        [Big Slide]      [YouTube] Chrome Click!!
Failure perception in UW-LFRR is fully passive and local. A node detects failures through missing ACKs or failure information carried in packet headers, and it maintains a dynamic failure set that accumulates both newly observed and previously known failures. Throughout this process, there is no need for probing packets and no need for broadcasting, which avoids additional control overhead on the acoustic channel.
ICACT20260346 Slide.12        [Big Slide]      [YouTube] Chrome Click!!
In a shortest path tree, when a node fails, it effectively severs its entire downstream subtree. UW-LFRR captures this with a branch-off abstraction: for each failed node, it defines the set of downstream nodes whose tree paths must traverse that failed node, and then unions these sets into a failure-affected region Ku. The key role of Ku is to turn what looks like a network-wide problem into a localized subgraph problem, so subsequent relay selection and searching are confined to Ku rather than the whole network.
ICACT20260346 Slide.13        [Big Slide]      [YouTube] Chrome Click!!
This abstraction makes the locality intuition very clear: the failure primarily impacts the downstream branch that depended on it in the shortest path tree, not the entire network. Other areas may still be adjacent to the failed node, but they do not depend on it to reach the destination, so they do not need to participate in repair. By focusing only on Ku, UW-LFRR avoids unnecessary computation and prevents routing state from spreading beyond the truly affected region.
ICACT20260346 Slide.14        [Big Slide]      [YouTube] Chrome Click!!
Before computing any detour, UW-LFRR performs topology purification. It defines a forbidden set that includes both the failed nodes and the blocked branch nodes in Ku, then removes them from the snapshot to form a local purified topology. This purified view ensures that the logical paths computed afterward are composed only of healthy, reachable nodes, so the repair will not point packets into dead ends.
ICACT20260346 Slide.15        [Big Slide]      [YouTube] Chrome Click!!
A single recovery strategy is usually unable to satisfy both speed and connectivity under multi-node failures, so UW-LFRR uses a two-stage design. Phase I prioritizes quickly returning traffic to the original shortest-path backbone whenever possible, and Phase II is invoked only when the backbone is locally cut and connectivity must be preserved through a constrained search inside the affected region. This follows a safety-first, fallback-as-needed philosophy.
ICACT20260346 Slide.16        [Big Slide]      [YouTube] Chrome Click!!
The goal of Phase I is to rejoin the original shortest-path backbone as quickly as possible. Within Ku, it identifies healthy nodes that still lie on the original shortest path to the destination, then finds a suitable cross-boundary relay pair so traffic can make a short hop across the boundary and graft back onto the backbone. Once traffic has rejoined the backbone, repair stops immediately and subsequent forwarding continues along the original shortest path, because the pre-failure shortest path is already globally optimal and stable; what we need is only a temporary bypass around a small failure region.
ICACT20260346 Slide.17        [Big Slide]      [YouTube] Chrome Click!!
Phase I is attractive because the detour is tightly controlled, stability is high, the amount of temporary state is minimized, and the computation remains low. In fact, this backbone-oriented search is designed to run with locality-bounded complexity on Ku, and it provides a bounded path stretch, so the repair avoids excessive detours.
ICACT20260346 Slide.18        [Big Slide]      [YouTube] Chrome Click!!
If Phase I cannot graft traffic back to the backbone, that indicates the backbone is completely disrupted in the local area. In that case, Phase II relaxes the backbone-first constraint and runs a shortest-path search on the purified topology restricted to Ku. The emphasis here is not on global optimality, but on guaranteeing that connectivity is preserved whenever the physical topology still allows a path to exist.
ICACT20260346 Slide.19        [Big Slide]      [YouTube] Chrome Click!!
Phase II provides two important guarantees. As long as the network remains physically connected, the constrained search on Ku will not miss an available route, and because the search space is restricted to Ku, its computation and energy cost are far lower than running a global shortest-path computation on the entire network. Loop-freedom follows from how the repaired segments are constructed. Each repair segment either follows a tree path in a shortest path tree or follows a shortest path on the purified topology, making it loop-free by construction. During recursive upgrades under multiple failures, the failure set only grows, newly failed nodes are pruned out of the view, and temporary entries have finite lifetimes with a latest-first selection rule, so outdated rules do not persist and packets do not get trapped in stable loops.
ICACT20260346 Slide.20        [Big Slide]      [YouTube] Chrome Click!!
After a repair path is computed, it must take effect immediately. UW-LFRR uses a first-packet paving mechanism: the first packets affected by a failure carry a compact control triplet, and as they travel, they install the necessary temporary forwarding entries along the detour. After that, subsequent packets can bypass the damaged area transparently without carrying extra control information and without triggering further computation.
ICACT20260346 Slide.21        [Big Slide]      [YouTube] Chrome Click!!
These temporary entries are strictly local soft state, scoped to a destination and failure context, and each entry has a finite TTL so it expires automatically. When multiple failures occur, a node can keep multiple entries for the same destination under different failure contexts, and the entry with the largest remaining TTL always takes precedence, ensuring the most recent failure view overrides older paths. With local-only installation, bounded lifetimes, and a latest-first policy, the temporary state remains confined near the failure and naturally fades as normal routing converges.
ICACT20260346 Slide.22        [Big Slide]      [YouTube] Chrome Click!!
In an example execution, the first affected packet triggers repair, detours around the failure region, and quickly rejoins a healthy path. Subsequent packets then follow the repaired route transparently and do not trigger additional computation. In other words, repair happens once on demand, and the effect persists until the control plane eventually converges and the network returns to its steady state.
ICACT20260346 Slide.23        [Big Slide]      [YouTube] Chrome Click!!
If the repaired path encounters a new failure, the node that first detects the new failure promotes itself to a new repair node, extends the accumulated failure list, rebuilds its local purified view, and reruns the two-stage selection to compute a new detour. This turns a complex multi-failure event into a chain of local repairs, and because the failure set grows monotonically and the view is pruned accordingly, the search region effectively tightens, avoiding repeated oscillations under the same stale rule set.
ICACT20260346 Slide.24        [Big Slide]      [YouTube] Chrome Click!!
We implemented UW-LFRR in NS-3.40 with a realistic underwater acoustic PHY and a TDMA-based MAC. The simulation uses a 49-node grid over a 10000 m × 10000 m area, with 4 kbps effective bandwidth and 1500 m/s sound speed. After the network converges, we randomly mark N nodes as unavailable, where N ranges from 0 to 20, to emulate sudden multi-node failures, and each configuration is repeated 100 times using a Monte Carlo method. The evaluation focuses specifically on the convergence-blind window before routing updates take effect, and the baselines include HL-OLSR, RB-AODV, FRP, FCP, and RTR.
ICACT20260346 Slide.25        [Big Slide]      [YouTube] Chrome Click!!
We report packet delivery ratio and end-to-end delay under increasing numbers of failed nodes. As failures grow, many schemes see delivery ratios drop quickly, while UW-LFRR maintains strong resilience across traffic patterns, and in all-to-all and random-burst scenarios its delivery ratio closely tracks the theoretical upper bound set by flooding-based protocols. On delay, schemes that repeatedly rediscover routes or activate backups on slow acoustic channels suffer large delays, while UW-LFRR maintains moderate and stable delay while still sustaining high delivery, which shows that localized, data-plane repair can effectively rescue critical traffic during CBZ.
ICACT20260346 Slide.26        [Big Slide]      [YouTube] Chrome Click!!
Putting the results together, UW-LFRR achieves high delivery with low and stable end-to-end delay, and it does so with locality-bounded computation rather than heavyweight repeated recomputation. The overall picture is that it may not be the most extreme winner on any single metric, but in a resource-constrained underwater setting, the balance between reliability, delay, and resource usage is exactly what makes it practical to deploy.
ICACT20260346 Slide.27        [Big Slide]      [YouTube] Chrome Click!!
We further examine control and computation overhead. Broadcast-driven schemes can exhibit prohibitive overhead—often exceeding 80% due to broadcast storms—while UW-LFRR incurs negligible overhead that is virtually identical to the HL-OLSR baseline, because signaling is embedded in data headers and state is restricted to the failure locality. In terms of computation, FCP can incur significantly higher cost because it triggers recomputation at each informed node along the route, whereas UW-LFRR computes and installs the detour upon the first packet and allows subsequent packets to bypass the failure without further computation.
ICACT20260346 Slide.28        [Big Slide]      [YouTube] Chrome Click!!
Overall, UW-LFRR demonstrates strong resilience during convergence-blind windows caused by sudden multi-node failures. Its data-plane repair suppresses blackhole propagation and maintains near-normal delay before control-plane updates arrive, and compared to alternatives that suffer from stale-topology forwarding, probing overhead, invalid backup activation, or heavy per-hop recomputation, UW-LFRR provides the most balanced combination of reliability, efficiency, and resource usage among the evaluated mechanisms.
ICACT20260346 Slide.29        [Big Slide]      [YouTube] Chrome Click!!
To summarize the contributions, we introduce a two-stage relay selection strategy that operates strictly within the failure-affected branch-off region Ku, prioritizing a backbone-oriented fast search and using a branch-region search only when necessary, which balances bounded path stretch and guaranteed connectivity while reducing computation. We also propose a snapshot-based local recovery framework with first-packet paving to install temporary, loop-free forwarding entries using only local information, eliminating network-wide flooding and minimizing control overhead. Finally, we provide analytical guarantees for loop-freedom and efficiency, and validate the performance gains through NS-3 simulations across diverse multi-failure scenarios.
ICACT20260346 Slide.30        [Big Slide]      [YouTube] Chrome Click!!
In closing, UW-LFRR is a purely data-plane, localized, loop-free fast recovery mechanism designed specifically for the most fragile period in underwater networking: the convergence blind zone. Rather than pursuing global optimality during this window, it ensures that packets can still be delivered, delay remains controllable, and energy is not squandered by excessive control traffic or heavyweight recomputation. Looking forward, the same foundation can be extended toward deeper cross-layer optimization, validation beyond simulation, and more complex underwater deployments such as coordinated AUV formations and 3D topologies.