Jul 2026· International Conference on Computer Communications and Networks· pp. 1-9· 0 citations· 26 references
Abstract
The built-in caching capability of Named Data Networking (NDN) is one of the most transformative proposals of next-generation network architecture, as it simultaneously realizes network traffic reduction, resilience to node failures, and prompt data retrieval. However, existing caching policies either make too many caches across the network by naively copying content everywhere or incur excess communication overhead for efficient caching through cache state exchange. Hence, we propose a machine learning-based caching policy along with a communication-efficient method to share the cache state. As the intelligent caching policy is capable of learning request patterns and estimating cache states of other nodes, our proposal realizes a more efficient utilization of the cache capacity by virtually considering the cache spaces of neighboring NDN nodes as an aggregated, larger cache space. The extensive evaluation experiments demonstrate the effectiveness of our proposal in increasing the cache hit ratio and reducing the average distance that each data travels to reach a requesting user. The effectiveness of the proposed cache state sharing method is empirically verified through an interpretable machine learning technique.
The rapid growth of internet users has significantly increased data traffic, demanding efficient content distribution architectures. Named Data Networking (NDN), an implementation of Information-Centric Networking (ICN), addresses this through in-network caching at router nodes. However, existing NDN caching strategies tend to address either cache placement or cache replacement in isolation. Placement strategies such as PoolCache effectively eliminate intra-cluster content redundancy through cooperative caching, but rely on standard Least Recently Used (LRU) replacement. Replacement strategies such as Least Recently and Frequently Used (LRFU) improve cache utilization through Combined Recency-Frequency (CRF) scoring, but without collaborative placement coordination. This paper proposes an integrated caching strategy that combines PoolCache and LRFU to simultaneously address both challenges. PoolCache manages caches across neighboring nodes as a shared resource, assigning each content to exactly one node per cluster via consistent hashing without extensive inter-node communication. LRFU complements this by evicting content with the lowest CRF score, retaining content that is both recently and frequently accessed. The integration is implemented and evaluated in ndnSIM 2.7 under two scenarios: variation of Content Store (CS) deployment percentage (5%-100%) and consumer demand rate (5, 20, 50, and 100 pkt/s) with Zipf-Mandelbrot content distribution $(\alpha=0.7)$. Results show that PoolCache+LRFU consistently outperforms PoolCache+LRU across all evaluated conditions. Under the CS percentage scenario, PoolCache+LRFU achieves a hit ratio up to 0.3666 compared to 0.3574 for PoolCache+LRU at $\mathbf{C S} \boldsymbol{=} \mathbf{1 0 0} \boldsymbol{\%}$, with the largest gap at $\mathbf{C S} \boldsymbol{=} \mathbf{2 0} \boldsymbol{\%}(\mathbf{0. 1 7 4 5}$ vs. 0.1375). Under varying demand, PoolCache+LRFU maintains stable average retrieval latency of 0.0318 s at 100 pkt/s, whereas PoolCache+LRU saw a performance decline to 0.0470 seconds, representing a 32.3% increase. Content redundancy is effectively suppressed by PoolCache in both configurations.
Muhammad Farhan, L. V. Yovita, Istikmal· International Seminar on Int...· 0 citations
Walking down the garden path, Named Data Networking (NDN) is a new networking approach based on the concept of content retrieval; it uses a unique name for a content request rather than retrieving data by host location (i.e., IP address-based retrieval). For NDN, timely and accurate caching of popular content along the path between producer and consumer will be a vital condition. The main concept of NDN is that each router stores data to quickly fulfil consumer requests; this is known as caching. Many cache replacement algorithms have been introduced to improve cache utilisation when the cache becomes full. This paper proposes a pending interest table-based cache replacement, i.e., PIT-REPL, which evicts the content from memory based on least score and further cache hit ratio enhance by round trip time based caching i.e. RTT-Cache and also developed a novel hybrid cache replacement strategy based on pending interest table and Q-Learning i.e. PITQ-REPL strategy to evict the content from memory based on cumulative reward which derive from score of pending interest table and Q-value reward of Q-learning. We found that the PIT-REPL cache hit ratio is enhanced by RTT-Cache-based cache replacement, i.e., RTT-Cache-PIT-REPL, and further cache hit ratio is enhanced by a hybrid approach of PIT and Q-learning strategy, i.e.,PITQ-REPL. Finally, we simulated using the ndnSIM simulator and found that the PIT-REPL replacement policy alone yields suboptimal cache hit ratios. However, the combined strategy of RTT-Cache with PIT-REPL enhances the cache hit ratio and reduces the latency, and finally hybrid strategy of PIT-REPL with Q-Learning significantly improves the cache hit ratios of RTT-cache-PIT-REPL and decreases the Latency up to some extent. Finally, the results show that the PITQ-REPL is a superior cache replacement to both strategies, namely RTT-cache-PIT-REPL and PIT-REPL.
Unknown authors· Global Journal of Engineerin...· 0 citations
Named Data Networking (NDN) enables in-network caching as a native mechanism to improve content delivery
efficiency and reduce network overhead. Cache replacement is an important factor for improving NDN performance.
However, most existing cache replacement strategies are derived from traditional host-centric systems and primarily rely
on content recency or insertion order, while largely ignoring NDN-specific state information. In particular, the Pending
Interest Table (PIT) and Forwarding Information Base (FIB) maintain valuable real-time knowledge about content
demand dynamics and forwarding behaviour, which remains underexploited in caching decisions. This paper proposes
two approaches: (i) a dynamic cache replacement strategy, PIT-DCREPL, based on parameters of the PIT table, and (ii) a
selective caching technique, PIT–FIB–aware selective cache. The proposed schemes integrates PIT metadata and FIB
forwarding information to enhance the CHR ( cache hit ratio) by adding a selective cache that prioritizes content with
higher expected reuse and forwarding importance. The simulations are conducted using the ndnSIM simulator, varying
the content store sizes where the performance evaluation demonstrates that PIT–FIB–aware caching strategies
consistently increase the cache hit ratio in cache replacement algorithms. The results confirm that leveraging PIT and FIB
information enables adaptive, intelligent, selective caching behaviour in NDN, thereby improving cache efficiency and
overall network performance.
S. Bagi, Neeraj Kumar· International Journal of Com...· 0 citations
Coded caching under various heterogeneous settings has been intensively studied. Specifically, in a shared-link model, a coded caching scheme based on a popularity-first caching strategy has been proved to be order-optimal under arbitrary popularity distributions. It is of great interest to investigate whether a similar conclusion holds in a device-to-device (D2D) network. Unlike shared-link models, since the server does not participate in the delivery phase in D2D networks, the caching scheme must ensure that all files can be recovered from the union of users’ caches. In this paper, we consider a D2D network constrained to uncoded prefetching and propose a new achievable scheme that caches popular files preferentially while allocating the minimal but necessary cache space to unpopular files. Then we derive new lower bounds on the average transmission rate under arbitrary popularity distributions. The proposed lower bounds are established via a novel decoupling method that isolates the rate analysis of popular files from the cache space allocated to unpopular ones, which could be of independent interest for future work. Finally, we prove that our proposed scheme achieves order-optimality among uncoded schemes under arbitrary popularity distributions. Numerical results validate the effectiveness of the proposed scheme and the tightness of the new lower bounds.
Jinbei Zhang, Wenjie Guan, Kai Huang et al.· IEEE Transactions on Communi...· 0 citations
In-network caching allows Named Data Networking (NDN) to inherently lower latency and bandwidth consumption, but it also creates prime vectors for security vulnerabilities like Cache Pollution Attacks (CPA) and Side-Channel Timing Attacks (SCTA). Conventional defenses typically treat security and content routing as isolated issues, applying filters reactively only after the cache has already been compromised. This paper introduces Trust-Weighted Proactive Caching (TWPC), a protocol designed to stop malicious content from ever reaching the Content Store by embedding dynamic trust evaluation directly into the packet forwarding pipeline. For every incoming request, TWPC aggregates content demand, name-prefix history, and immediate requester anomalies to generate a single Trust Score. Suspicious Interests are preemptively dropped or explicitly flagged so as to bypass storage. Rather than relying on static thresholds, an on-line reinforcement learning loop dynamically recalibrates policy parameters to match shifting attack patterns. We evaluated our protocol across multiple topologies (including GEANT and Abilene) using ndnSIM. Under active CPA and SCTA stress, TWPC consistently maintained a 22.6% higher cache hit ratio and saved 19% in overall latency compared to standard baselines, achieving a 97.4% detection accuracy.
Unknown authors· International journal of com...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
