Source-linked AI summary
Information Centric Networking in the IoT: Experiments with NDN in the Wild
Emmanuel Baccelli, Christian Mehlis, Oliver Hahm, Thomas C. Schmidt, Matthias Wählisch
TL;DR
The paper asks whether ICN’s benefits outweigh its shortcomings in constrained IoT deployments, where memory, energy, bandwidth, and unreliable connectivity matter. It evaluates NDN in a life-size deployment, introduces interoperable routing and caching enhancements, and reports lower resource use, reduced transmissions, and improved availability, while identifying wireless-link and scaling limitations.
Problem
It remains open whether ICN’s benefits outweigh its shortcomings in IoT, particularly under constrained memory and routing requirements.
Method
The paper conducts NDN experiments on a 60-node real IoT deployment and evaluates interoperable enhancements involving routing, data paths, and caching.
Results
ICN reduced ROM size by up to 60%, RAM size by up to 80%, radio transmissions by about 50% with RONR, and by up to 50% with caching.
Takeaways & Limitations
The experiments show that ICN is applicable in IoT and can offer energy, memory, ROM, routing-state, and content-availability advantages through interoperable enhancements.
Takeaways & Limitations
IoT wireless links can be asymmetric, unidirectional, and variable over time, constraining ICN assumptions and deployment conditions.
Abstract
from arXiv · showhide
This paper explores the feasibility, advantages, and challenges of an ICN-based approach in the Internet of Things. We report on the first NDN experiments in a life-size IoT deployment, spread over tens of rooms on several floors of a building. Based on the insights gained with these experiments, the paper analyses the shortcomings of CCN applied to IoT. Several interoperable CCN enhancements are then proposed and evaluated. We significantly decreased control traffic (i.e., interest messages) and leverage data path and caching to match IoT requirements in terms of energy and bandwidth constraints. Our optimizations increase content availability in case of IoT nodes with intermittent activity. This paper also provides the first experimental comparison of CCN with the common IoT standards 6LoWPAN/RPL/UDP.
1. INTRODUCTION
The paper examines whether ICN can address IoT networking needs despite constrained devices, incomplete prior evidence, and unresolved trade-offs. It reports real-world NDN experiments and proposes interoperable enhancements evaluated against established IoT protocols.
- ICN is investigated as a networking approach for IoT’s large-scale machine-to-machine and named-content communication setting.
- ICN’s potential benefits include cache-assisted replication and reduced complexity, while freshness, unscheduled actuator traffic, and memory growth remain challenges.
- The paper frames its objective as experimentally exploring whether ICN’s benefits outweigh its IoT shortcomings.
- Prior work lacked large-scale experiments on constrained IoT hardware in realistic deployments and often targeted MANETs or sink-centric sensor traffic.
- The paper reports CCN experiments in a life-size building deployment and compares them experimentally with 6LoWPAN/RPL/UDP.
- The proposed interoperable enhancements reduce interest traffic, use data paths and caching, and increase content availability for intermittently active nodes.
2. A PRIORI CHALLENGE OF ICN IN IOT: LIMITED MEMORY
Limited memory is a fundamental constraint for applying ICN to IoT devices. The section examines caching, protocol-stack, and routing requirements, identifying designs that can fit constrained hardware while retaining ICN benefits.
- Constrained IoT devices leave very little RAM available for ICN content caching after operating-system, network-stack, and application requirements.Available RAM is usually around 10 kBytes, with remaining cache capacity at most around 1 kByte.
- Caching remains feasible because temperature readings are small and distributed strategies can coordinate storage of larger content across devices.A 12-byte temperature value allows approximately 85 sensor values in one cache.
- Multiple consumers and intermittent node activity make caching useful by improving content availability and shortening lossy paths to producers.
- Running ICN above IP can be impractical because additive ICN and IP memory requirements exceed constrained-device capabilities.
- Proactive link-state routing conflicts with IoT constraints through ongoing control traffic and memory requirements typically growing as O(n).
- The proposed routing direction targets O(1) routing state and minimal control traffic, ideally none when no data traffic exists.
3. STEPS TO ENABLE ICN IN THE IOT
The paper enables ICN experimentation on constrained IoT hardware by porting a compact NDN implementation to RIOT and deploying it across a realistic wireless testbed. It also addresses configuration, packet sizing, and resource-footprint constraints.
- CCN-Lite was ported to RIOT because NDN can operate directly above the link layer and CCN-Lite is compact and reference-compatible.The implementation uses less than 1,000 lines of C code and has a low memory footprint.
- NDN and 6LoWPAN/RPL stacks were compared on contemporary IoT operating systems and hardware using ROM and RAM footprint measurements.
- ICN reduced ROM size by up to 60% and RAM size by up to 80% relative to the compared common IoT protocols.
- The deployment contains 60 wireless nodes distributed across rooms, floors, and buildings, matching building-automation density and environmental conditions.
- FIB autoconfiguration is required because IoT deployments cannot rely on humans for manual configuration or universally predefined location-based naming.
- Names and chunk sizes must fit roughly 30–100-byte IoT link-layer MTUs because ICN fragmentation is not addressed.
4. NDN EXPERIMENTS AND OPTIMIZATIONS FOR IOT DEPLOYMENT
The experiments deploy NDN across a 60-node, multi-floor IoT testbed and evaluate routing, caching, and comparison with 6LoWPAN/RPL/UDP. RONR and caching reduce transmissions relative to interest flooding, while NDN uses fewer transmissions than the compared IoT stack.
- 4.1 Large-scale Deployment Setup: A 60-node NDN deployment spans rooms, floors, and buildings, with nodes operating autonomously over wireless links.Docking stations support monitoring and management but are not used for node communication.
- 4.2 Vanilla Interest Flooding (VIF): Each 58-byte chunk carries 30 bytes of content without fragmentation, and consumers periodically fetch 10 chunks.The configuration matches the 64-byte link-layer MTU used in the deployment.
- 4.2 Vanilla Interest Flooding (VIF): VIF requires k · ((n −1) + √n) transmissions for k chunks in an n-node network because each chunk triggers network-wide interest flooding.This transmission growth does not scale well as network or content size increases, and radio communication is costly for battery-powered devices.
- 4.3 Reactive Optimistic Name-based Routing (RONR): RONR floods only the first interest and installs temporary reverse-path FIB entries so subsequent chunk interests are unicast.In the same topology and scenario, RONR reduces radio transmissions by about 50% compared with VIF and scales better as network or content size grows.
- 4.4 Multiple Consumers & Impact of Caching: With 20-chunk caches on all nodes, transmissions decrease by up to 50% because cached chunks shorten unicast paths.The cache uses 2 kBytes, or 2% of the 96-kbyte overall RAM, and broadcast transmissions remain almost similar.
- 4.5 Comparison with 6LoWPAN/RPL/UDP: 6LoWPAN/RPL/UDP produces approximately three times more transmissions than NDN in the corresponding multi-consumer experiment.The paper attributes the gap mainly to proactive control traffic and paths that do not benefit from caching.
5. A POSTERIORI CHALLENGES: WHAT ARE THE LESSONS LEARNED
The experiments expose energy, bandwidth, memory, wireless-connectivity, and communication-model challenges for ICN in constrained IoT networks. The proposed mechanisms use shorter names, less flooding, caching, and local multicast to address these constraints, while several deployment limitations remain.
- Energy consumption: k chunks requested through VIF generate k · ((n −1) + √n) transmissions because each chunk triggers a network-wide flooded interest.Although NDN operated on the IoT hardware and fetched content, many packets were transmitted relative to the content size.
- Energy consumption: Flooding costs O(n) packet transmissions and O(nm) packet receptions, making it expensive for constrained IoT devices.Received packets also incur processing costs, with string comparison and name matching representing 85% of measured CPU cycles in the RIOT implementation.
- Energy consumption: Short hierarchical names reduce fragmentation and leave roughly 30 bytes for payload with 12-byte names and a 64-byte link-layer frame.A deeper example name, /zone1/room2/dev7/temp/a, leaves about 20 bytes for content payload.
- Energy consumption: Human-readable or cryptographically generated names may waste constrained IoT resources, motivating compact naming and CCN header compression.The paper does not consider security-related naming aspects, and security mechanisms typically produce longer headers.
- Energy consumption: Even small in-network caches can reduce energy use by shortening paths and allowing content producers, such as sensors, to sleep while cached content remains available.The paper notes that caching can reduce the number of intermediate energy-challenged devices involved in retrieving content.
- Wireless connectivity: CFA sends one link-local multicast for multiple matching consumers, while ONPC caches overheard chunks to improve availability and reduce later radio transmissions.CFA avoids explicit location awareness, but the paper reports no experimental results for CFA or ONPC because of space constraints.
- Wireless connectivity: IoT wireless links have small frames and may be asymmetric or unidirectional, with link quality varying significantly over time.The paper reports examples of 10% loss in one direction and 80% in the reverse, as well as links that work in only one direction.
- Wireless connectivity: Routing must combine negligible control traffic and small state with dynamic link tracking to avoid forwarding interests over failing reverse paths.Poor reverse paths can cause failed content delivery, PIT timeouts, and additional interest flooding.
6. CONCLUSION AND PERSPECTIVES
The paper demonstrates NDN on constrained IoT hardware in a life-size deployment and uses the experience to assess ICN's applicability and advantages. It concludes that ICN can improve energy use and memory footprints relative to 6LoWPAN/IPv6/RPL, while motivating interoperable enhancements.
- Conclusion and perspectives: Experiments deployed NDN on tens of constrained nodes across multiple rooms and buildings, demonstrating that ICN is applicable in the IoT.The deployment was life-size rather than a small toy network.
- Conclusion and perspectives: The paper reports advantages over 6LoWPAN/IPv6/RPL in energy consumption and RAM and ROM footprint.This conclusion is based on experiments with NDN in the real IoT deployment.