Expected Forwarding Delay for Latency Aware Routing

22 concepts16 papers1 proposal

Traditional routing protocols often rely on static or topology-centric metrics, such as minimizing hop count or maximizing raw signal strength. While intuitive, these metrics treat network nodes as transparent pipes, ignoring a fundamental reality: a packet does not just travel across physical space; it travels through time.

In real-world networks, a packet spends a significant portion of its lifespan waiting inside nodes due to queuing, medium contention, node sleep schedules, and coordination overhead. To minimize end-to-end latency, a growing body of research focuses on Expected Forwarding Delay (EFD) and latency-aware path selection. By modeling and predicting the exact temporal costs incurred at each hop, these techniques allow networks to route around invisible, dynamic bottlenecks.

Decomposing the Internal Delays of a Node

To understand why traditional metrics fail, consider a node that is physically close to the destination but buried under a mountain of traffic. A hop-count metric will eagerly send packets to this node, where they will sit in a buffer or repeatedly fail to clear the medium.

To address this, Islam et al. introduced the Expected Forwarding Delay (EFD) metric for wireless mesh networks. EFD shifts the focus from link quality alone to a holistic calculation of the time a packet spends within a node before successful transmission. This value aggregates:

  1. Queuing Delay: The time a packet waits in line behind other traffic.
  2. Medium Access Delay: The time spent waiting for the wireless channel to become free, including backoff and deferral times dictated by network contention and traffic priority.
  3. Transmission Time: The actual physical duration required to push the bits across the link.

Similarly, Chen et al. proposed Expected Transmission Delay (ETD), which incorporates MAC-layer handshaking times and bidirectional success rates to quantify link traversal duration accurately. Rather than relying on raw signal strength, Athanasiou et al. utilized an "airtime metric" that captures protocol overhead and frame error rates, exposing the true capacity consumed by a wireless link.

In wired or software-defined architectures, this internal delay can be modeled mathematically or measured via infrastructure. For instance, Son applied an M/M/1 queue model within a Software-Defined Network (SDN) to predict average waiting times on potential paths, allowing the controller to proactively steer traffic away from building congestion. Taking a measurement-based approach, Kim calculated an Estimation Value of Transmission Delay (EVTD) by monitoring available bandwidth directly on switch ports, mapping flow characteristics to real-time path latencies.

The Temporal Costs of Opportunistic Forwarding

In low-power environments like Wireless Sensor Networks (WSNs), nodes frequently utilize duty-cycling—alternating between active and sleep states to conserve energy. In these environments, the primary source of delay is often not congestion, but the time spent waiting for a target neighbor to wake up.

Traditional unicast routing forces a node to wait for one specific neighbor, creating a massive latency penalty. Opportunistic routing breaks this paradigm by allowing a node to forward a packet to the first available neighbor out of a set of candidates.

Unicast Routing:
[Node A] ----(Waiting for Node B to wake up)----> [Node B (Asleep)]

Opportunistic Routing:
                  /---> [Node B (Asleep)]
[Node A] --------+----> [Node C (Wakes up first! Receives packet)]
                  \---> [Node D (Asleep)]

To accurately capture this dynamic, Landsiedel et al. designed a metric that calculates the expected number of wakeup intervals needed based on the joint probability of any node in a candidate set waking up. By expanding the selection pool, the metric reflects the drastic latency reductions inherent in opportunistic forwarding. This strategy is echoed by Duan et al., who demonstrated that restricting next-hop choices to a dynamically optimized set of high-quality candidates minimizes waiting times and slashes end-to-end delay.

However, opportunistic routing introduces a hidden penalty: coordination delay. When a node broadcasts a packet to multiple candidates, those candidates must coordinate so that only one of them forwards it further. If a lower-priority candidate acts too quickly, it might duplicate the packet or steal the route from a better positioned but slightly slower neighbor.

As Cheng et al. observed, failing to account for this coordination wait time leads to suboptimal path selection. The authors demonstrated that the temporal cost of relay selection is just as critical as physical link quality; their metric explicitly balances physical transmission time with the coordination overhead required to confirm that a higher-priority neighbor has not already acted.

Predictive Modeling and Multi-Path Expectations

Because network conditions fluctuate rapidly, modern routing frameworks rely on predictive modeling and multi-path optimization to estimate forwarding delays before a packet is ever sent.

In highly dynamic environments, such as intertidal zones or Flying Ad-Hoc Networks (FANETs), topology changes are severe. Zhou et al. addressed this by combining predicted node availability with link quality to map out paths with the lowest anticipated end-to-end latency. For highly mobile UAV networks, Xue et al. leveraged Q-learning to dynamically incorporate predicted end-to-end delay into routing value calculations, ensuring paths remain functional and efficient over time. In fog computing environments, Li et al. bypassed complex active probing by using virtual geometric coordinate systems to predict inter-node latency with high accuracy.

When multiple paths are available, calculating delay requires shifting from single-path metrics to probabilistic multi-path expectations. In Delay Tolerant Networks (DTNs), where node encounters are highly sporadic, Jeong et al. introduced the ExMin metric. Instead of evaluating the minimum delay of individual paths in isolation, ExMin calculates the mathematical expectation of the minimum delay across all possible routes combined. This approach leverages the joint probability of multiple opportunistic encounters, recognizing that having three mediocre, independent paths can sometimes yield a lower expected delay than relying on a single "best" path.

This multi-path perspective applies to internet-scale and application-layer routing as well:

  • Onion Routing (Tor): Imani et al. maintained a small pool of pre-built circuits and used real-time Round Trip Time (RTT) measurements to select paths. RTT serves as a superior proxy for forwarding delay because it simultaneously captures both physical propagation distance and dynamic node congestion.
  • Voice-over-IP (VoIP): Gong et al. utilized a path diversity scheme that independently evaluates multiple network paths and dynamically selects the route that minimizes jitter buffer delay, balancing conversational latency with packet resilience.

Exploiting Delay for Protocol Efficiency

Remarkably, understanding forwarding delay allows researchers to turn latency into an optimization tool for the routing protocol itself. In distributed networks, routing updates can flood the network and consume precious energy and bandwidth.

Aslam et al. designed a protocol that intentionally introduces a variable propagation delay to routing updates, making the delay proportional to the path cost.

[Source] --(Low Cost Update)--> [Node A] --(Fast Forward)--> Arrives First!
[Source] --(High Cost Update)--> [Node B] --(Intentional Delay)--> Suppressed!

Because optimal routing messages experience minimal delay, they propagate through the network first. When a node receives an optimal update, it automatically suppresses any subsequent, suboptimal updates traveling along slower paths. By exploiting the physics of delay, the protocol naturally prunes control overhead without requiring complex, explicit coordination.

Key Foundational Work

High throughput path selection for IEEE 802.11s based wireless mesh networks (2010) — contributed 2 concept elements to this direction.
Distributed Energy-conserving Routing Protocols (2003) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_expected_forwarding_delay_eb9bcbce,
  title = {expected forwarding delay},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/expected-forwarding-delay/rdr_eb9bcbce11ae4fbdbabab2a73663f19a},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_expected_forwarding_delay_eb9bcbce_v1,
  title = {expected forwarding delay},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/expected-forwarding-delay/rdr_eb9bcbce11ae4fbdbabab2a73663f19a/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake