Controller-driven vector autoregression model for predicting content popularity in programmable named data networking devices

Software defined networking and information-centric networking

Software-defined networking (SDN) is an emerging network architecture that separates the control plane from the data-forwarding plane, enabling flexible control over network infrastructure (Mughees et al., 2023). Two main components play crucial roles in SDN-based content delivery networks, such as information-centric networking (ICN). The central entity, known as the SDN controller, manages and maintains the network topology to ensure the availability of various named contents received by participating CR routers. The second component involves a straightforward forwarding approach, with caching devices responsible for routing the requested packet along the path directed by the SDN controller (Ahad et al., 2017).

The integration of ICN into the SDN framework yields numerous advantages (Zhang et al., 2019; Mahmood et al., 2018). SDN’s intrinsic flexibility facilitates real-time dynamic routing adjustments, optimizing content delivery paths. It, coupled with centralized control, enhances content placement, effectively reducing latency and improving the overall user experience. Rigorous security measures, encompassing robust access control, encryption, and threat detection, are systematically implemented, ensuring a comprehensive and steadfast security stance.

Moreover, scalability enhancements have been introduced to cater to the growing demands on the network. Integrating emerging protocols and technologies ensures ongoing compatibility with evolving standards and innovative communication methods, highlighting the adaptability of the NDN framework. A pivotal benefit is evident in SDN-controller-aware caching within the NDN delivery system, leading to improved network routing and caching strategies (Mateen et al., 2023).

• (i)

  Routing: In SDN-based NDN delivery systems, rules, and routing decisions are vital for efficient data delivery. Centralized control is achieved by centralizing network management. When a user sends an INTEREST packet, the SDN controller programs the forwarding information base (FIB) table with routing rules derived from content names (Kalafatidis et al., 2022). These rules are strategically placed in the FIB table to map each content name to a specific route or set of routes. When an INTEREST packet is sent, content routers (CR) consult the FIB table to determine the best route for fulfilling the request.

• (ii)

  Caching: Caching in SDN-based NDN systems enhances data retrieval efficiency by strategically placing caches at network points like routers and switches. These caches store frequently accessed data, enabling swift data delivery upon request. Unlike traditional IP-based networks, SDN-based content-delivery networks cache data based on its unique name (Aldaoud et al., 2023). Its distributed caching approach reduces data duplication, minimizes network congestion, and fosters faster content delivery, ultimately enhancing overall network performance and user experience.

This article delves into popularity-based content placement in programmable NDN devices, employing a predictive time-series VAR model to anticipate content popularity trends. Concurrently, a heuristic algorithm, driven by an SDN controller and based on topological features, is proposed to identify popular placements for NDN devices. Additionally, this article aims to reduce miss rates by evaluating the local cache freshness of each received DATA packet, thereby avoiding the prefetching of frequently accessed content from external sources. Furthermore, to address data redundancy, a system of custom-named packets is implemented to prevent intermediate nodes from caching forwarded packets.

De-centralized caching schemes

Decentralized caching schemes have become prominent. Cache Everything Everywhere (CEE), which involves caching content at each router, increases redundancy while decreasing cache diversity (Chai et al., 2012). CacheFilter addresses redundancy, using a FlagBit to decide whether to cache or forward content (Feng et al., 2015). WAVE, a chunk-based caching scheme based on the content frequency and inter-chunk distance, was introduced by Cho et al. (2012). Probability Caching (PopCache) is a probability-driven scheme that caches content near consumers with high probability values and at a distance from low probability ones (Suksomboon et al., 2013). Furthermore, Random Caching (RC) utilizes 0 to 1 probability values for caching (Tarnoi et al., 2014).

Recent research has highlighted popularity-driven caching strategies. The push-down push-up (PDPU) policy dynamically moves content closer to users as its popularity increases, selecting cached content based on popularity (Nour et al., 2020). Another approach coordinates and shares popularity information among network routers, forming a popularity-aware caching system (Li et al., 2012). As presented in a different study, Most Popular Content (MPC) allows routers to cache content once it reaches a locally preset popularity threshold, with each router individually setting its threshold (Bernardini, Silverston & Festor, 2013). In popular content caching, Dynamic Fine-Grained Popularity-Based Computing (D-FGPC) is introduced as an extension of MPC. D-FGPC employs a flexible popularity threshold based on Interest frequency and cache capacity (Ong et al., 2014). Another strategy in the context of popularity-driven placement entails the implementation of a time-series autoregressive (AR) model (Liu et al., 2019). This approach enriches our understanding of content popularity dynamics by deliberately considering lagged orders associated with relevant variables.

As exemplified by CRCache and BEACON, Centrality router-based optimal placement schemes strategically leverage content popularity and key routers to optimize content placement for efficient delivery (Wang et al., 2014; Xiaoqiang, Min & Muqing, 2016). As demonstrated in Zheng et al. (2019), collaborative caching schemes consider content popularity and router betweenness centrality. It involves storing popular content on highly central routers and less popular content on routers with lower centrality. Another placement scheme is introduced, making popularity and placement decisions based on the exponential weighted moving average (EWMA) and the mathematical derivation of node centrality (Liu et al., 2021). Despite its effectiveness, the method encounters delays associated with the cache’s high proximity to users.

To address the challenge, Popularity-Aware Caching (PaCC) enhances content proximity to consumers by incorporating hop-distance awareness in placement decisions (Amadeo et al., 2022). However, hop-distance-based placement entails inefficiencies for large networks. Another method presented in Dutta et al. (2022) makes caching decisions based on node distance cost and congestion cost to optimize content placement decisions. Nevertheless, it suffers from suboptimal performance in retrieval delays due to its high computational cost. Subsequently, the Caching Popular Fresh Content (CPFC) strategy is introduced to reduce miss rates and retrieval delays, dynamically assessing both popularity and content freshness for cache suitability (Amadeo et al., 2020). In the context of caching policies for edge and core routers, as discussed in Amadeo et al. (2021), the approach prioritizes placing the most popular content at the edge and popular, long-lasting content on core routers.

The existing research on decentralized-based cache placement schemes has made valuable contributions, but has certain limitations and gaps. Some schemes utilize probabilistic caching based on content request rates, potentially resulting in suboptimal cache hit rates due to fixed probability thresholds. Others rely on popularity-driven metrics, often leading to reduced cache hit rates by focusing on a single factor, such as frequency rate. Some schemes incorporate betweenness centrality for cache placement, but it may cause delays in content retrieval, especially when the content provider is distant from consumers. Additionally, specific caching criteria consider popularity and non-transient freshness. However, no mechanism exists to evaluate each cached content’s local residual freshness period, leading to cache miss rates. Table 1 summarizes the objectives and limitations of the different cache placement schemes.

Centralized caching schemes

Researchers propose a central-entity-based caching approach within Information-Centric Networks (ICN) to optimize resource management. An autoencoder-based model predicts content popularity by analyzing spatial-temporal data collected by the SDN controller. This model empowers routers to employ a softmax classifier, selecting crucial nodes based on betweenness centrality to strategically place popular content (Liu et al., 2017b; Mughees et al., 2021). In the article (Narayanan et al., 2018), the authors employed the Long Short-Term Memory (LSTM) model for popularity prediction, but the lack of a strategically implemented placement approach results in heightened retrieval latency.

In recent work (Dudeja et al., 2022; Liu et al., 2022), an SDN-based secure content caching and forwarding approach was introduced. Routers maintain an index table in their cache memory with unique signatures for data. However, it is noteworthy that the implemented placement strategy for content caching is based on frequency count, which may not be sufficient to capture the changing trends over time, resulting in heightened delays in content retrieval.

While centralized-based caching in networks has utilized topological characteristics like betweenness centrality for content placement, this can result in longer content delivery times due to central nodes being distant from consumers. Recent research introduced closeness-aware metrics based on hop counts, which may increase computational and delivery costs without accounting for variable link capacities and latency. Some studies implement popularity-aware caching across multiple autonomous systems, basing caching decisions on request counts rather than consumer historical requests, potentially leading to suboptimal cache hit rates.

The limitations identified in the current body of research underscore the necessity for our study, aiming to bridge these gaps and present a comprehensive solution by introducing a novel caching policy. This research advocates for centralized popularity-aware placement in popular NDN nodes (PaCPn), where content popularity is calculated not solely based on request counts, as seen in existing studies, but also considers previous lag values for each requested packet to predict future popular content. Consequently, a multi-variant Vector Autoregression (VAR) popularity model is proposed, anticipating future popular content based on historical values of the frequency count and previous popularity scores.

To improve the efficiency of content delivery and minimize delivery times, we introduce a closeness-aware heuristic algorithm for caching popular content. This algorithm assesses node proximity through a metric divided into distance cost to consumers and content delivery time. Diverging from studies that rely on hop distance-based closeness, we leverage a content frequency rate to ascertain a node’s proximity to specific content. The preferred caching location is identified as a node with a high closeness metric for a given content.

Problems in existing cache placement strategies

• Suboptimal cache placement: Decentralized caching control has led to less-than-optimal cache placement decisions.

• One-factor-driven content popularity: Depending on frequency-based content, popularity may not fully capture the nuanced aspects of consumers’ behavior when requesting content.

• Popularity and freshness thresholds: Fixed thresholds for content popularity and freshness in caching decisions have resulted in suboptimal cache hit ratios.

• Weighted average-based popularity prediction: Inefficient cache miss rates are observed because past popularity records for specific content items are not considered in the weighted moving average used for popularity prediction across different time intervals.

• Centrality-metric-based placement selection: Choosing popular caching devices based on betweenness centrality can lead to high delays in content retrieval due to the distance between the source and content consumers.

• Hop-distance-based cache placement: Making cache placement decisions based on limited topological characteristics, including hop distance, has resulted in suboptimal content retrieval delays.

• Increased packet exchange in NDN: The exchange of traffic within caching devices has increased due to adding a “Flag_bit” in the DATA packet field, caching content near the producer but far from the consumers.

Objectives of the study

In summary, the objectives of this work aim to improve caching placement strategies:

• Programmable caching devices: Transform caching devices into storage units by implementing a centralized controller-driven caching system to manage caching within network devices.

• Multi-factor content popularity: Develop a caching approach that considers factors beyond content frequency, including historical popularity records and request counts, to predict content popularity more accurately.

• Time-series model for popularity prediction: Utilize satistical regression-based time series models to predict content popularity within discrete time intervals, thereby reducing cache miss rates by analyzing historical popularity trends.

• Topological-features-driven placement: Optimize the placement of popular content by considering topological characteristics like ‘Distance cost,’ ‘Delivery time,’ and ‘Content status.’

• Controlled traffic for NDN packets: Implement controlled traffic management using a customized DATA packet with a ‘Cacher’ field, allowing authorized sources to cache the packet while preventing intermediate nodes from caching it.

• Evaluating content freshness or validity: Calculate local freshness periods to reduce cache miss rates by piggybacking content onto requested sources.

Research contributions

This research article introduces a framework called ‘PaCPn,’ which utilizes an SDN controller to make informed caching decisions for NDN caching devices. This centralized caching system enhances network efficiency by dynamically managing caching resources and improving the transmission of NDN packets. Unlike complex decentralized caching decision strategies used in previous studies, this centralized approach offers a comprehensive solution that allows the network to quickly adapt to the requested content, optimizing resource utilization and reducing data redundancy. The research article addresses the following research questions:

• 1)

  How can the SDN controller be equipped with intelligence to periodically identify a set of popular contents for a content delivery network?

  • •

    Highly popular content is cached on edge devices using a multi-variant time series model (VAR) embedded within the SDN controller.

  • •

    The VAR model predicts popular content based on historical data, including ‘Frequency rates’ and ‘Popularity rates.’

• 2)

  What mechanisms does the SDN-controller employ to determine the set of popular nodes for caching the popular contents?

  • •

    PaCPn employs a closeness-aware heuristic algorithm guided by the SDN controller to place popular content strategically.

  • •

    Placement decisions consider factors like ‘distance cost’ (cache node proximity to consumers) and ‘delivery cost’ (time required to deliver content to designated consumers).

• 3)

  What strategy does the SDN caching network use to store packets with a valid residual freshness period when the DATA packet is piggybacked to the requested source?

  • •

    PaCPn improves traffic management by adding fields like ‘Cacher name’ to prevent data duplication and control traffic within caching devices during piggybacked content delivery.

• 4)

  What mechanism does the SDN system use to prevent intermediate nodes from caching a copy of content when communication occurs between the requested source and content producer?

  • •

    Caching devices utilize ‘Generation time’ and ‘Expected time’ fields to calculate each DATA packet’s local residual freshness period, ensuring content validity and freshness.

• 5)

  What replacement strategy is employed in this article?

  • •

    In cases of full caching space, less popular content is replaced with higher-priority content based on popularity values.

  • •

    This optimization improves cache hit ratios and reduces content delivery delays.

Decision parameters for content caching

The consumer’s requested content is strategically cached on edge devices based on specific decision parameters.

Proposed algorithms

The training process of the VAR popularity model, as outlined in Algorithm 1, follows a set of steps. In the first step (STEP-1), the algorithm initializes the coefficient matrix, representing the weights associated with the timestamp values of each variable. Iterative loops set the initial weights for the VAR coefficients, preparing the model for subsequent estimation.

In (STEP-2), the algorithm utilizes the ordinary least squares (OLS) technique to estimate the coefficients. It constructs lag matrices and employs OLS to update the coefficient matrix iteratively. This process educates the model about the relationships between past observations and future values.

In (STEP-3), the algorithm forecasts future content popularity using the trained VAR model. It employs the estimated coefficient matrix to predict popularity for the upcoming time steps. Iterative loops traverse forecast steps, leveraging the model’s understanding of temporal dependencies.

Finally, in (STEP-4), the algorithm evaluates the model’s efficiency by computing each variable’s root mean squared error (RMSE). RMSE assesses the alignment of predicted popularity values with actual values for the forecasted time steps, providing insights into the model’s accuracy and predictive capabilities.

Once the VAR model is trained, it is deployed for real-time predictions in a programmable content delivery network environment. Real-time predictions are generated by loading a pre-trained VAR model with coefficients and hyper-parameters. Data for the current timestamp, including relevant variables and past observations from the HT, is collected. Subsequently, predictions are made using the loaded coefficients matrix to estimate content popularity for the future timestamp.

The closeness-aware heuristic algorithm (Algorithm 2) optimizes content caching in a NDN environment. This algorithm takes as input the set of requested content C $r$ and the set of network cachers N $c$. It iterates through each requested content and computes a closeness metric $C_T[N_c[I],C_r[j]]$ for each caching device in the network. The closeness metric is calculated based on Eq. (4), resulting in a list for each requested content.

The algorithm then sorts these lists in ascending order using the insertion sort algorithm, effectively arranging caching devices regarding closeness to the requested content. Following this, the algorithm identifies the caching device with the highest closeness value (first item in the list) for each requested content. The chosen caching device and the associated content are inserted into the Popular Cacher List (PL).

In the subsequent iterations, the algorithm updates the Content Information Base (CIB) table by associating each caching device with its corresponding content, effectively populating the estimated popular nodes or NDN caching devices. This process continues until the Popular Cacher List is empty.

The repeated steps involve sending name prefixes with their corresponding caching device names to the content producer, ensuring that the requested contents are directed to their estimated popular caching devices. This iterative process continues until all requested contents are processed.

In summary, the closeness-aware heuristic algorithm efficiently selects the most suitable caching device for each requested content, optimizing the distribution of content across the NDN caching network. The algorithm’s reliance on closeness metrics enhances the effectiveness of content placement, contributing to improved cache hit rates and reduced content retrieval delays.

Processing of INTEREST packet at programmable NDN node

The INTEREST packet processing algorithm (Algorithm 3) in NDN involves crucial steps for handling incoming requests. When an INTEREST packet arrives, the algorithm conducts a lookup in the content store (CS) table. If a match occurs (CS hit), indicating that the requested data is cached, the algorithm searches for the corresponding DATA packet and promptly returns it to the intended interface.

In scenarios with no match in the CS, the algorithm checks the pending interest table (PIT) to determine if an entry already exists for the requested content (PIT hit). Upon a PIT hit, the algorithm updates the existing PIT entry by adding the incoming interface and awaits the corresponding DATA packet. Notably, the packet is not forwarded in this scenario, optimizing resource utilization.

Subsequently, if there is neither a match in the CS nor a PIT hit, the algorithm examines the forwarding information base (FIB) table. If a match is found in the FIB, the algorithm creates a new PIT entry for the requested INTEREST packet. Following this, the packet is forwarded based on the flow rule specified in the FIB.

When none of the CS, PIT, and FIB conditions are met, indicating a miss, the algorithm triggers an OF_Packet_IN message to the controller. The controller responds by installing a flow rule into the FIB using OF_FLOW_MOD, enabling the algorithm to forward the packet to the next station according to the FIB.

The proposed solution ensures effective request handling through CS, PIT, and FIB tables to enhance the efficiency of INTEREST packet management. It minimizes network latency by dynamically adapting to the current state. Integrating flow rules optimizes content retrieval or packet forwarding based on real-time conditions.

Figure 4 presents efficient INTEREST packet management utilizing CS, PIT, and FIB tables. It ensures effective request handling by retrieving cached content or forwarding packets based on the network’s current state and established flow rules, thereby contributing to a responsive and agile content delivery system.

Processing of DATA packet at programmable NDN node

Algorithm 4 outlines the processing of data packets in a NDN environment. Upon the arrival of a DATA packet, the algorithm performs a Pending Interest Table (PIT) lookup to determine the caching and forwarding decisions. If a PIT hit occurs, indicating an existing entry for the content in the PIT, the algorithm calculates the content’s freshness using Eq. (5).

The subsequent steps of the algorithm involve evaluating conditions based on the caching node’s name (c $n$) and the potential cacher name (p $cn$), along with the local freshness period $L_{(ck)}[T_{res}]$ within the NDN environment.

When the names of the caching node and the potential cacher match, and the freshness period is non-transient (indicating an extended period), the algorithm opts to cache the received DATA packet. This decision signifies that the content is fresh and can be stored within the cache for future requests. It indicates recognition by the caching node that it already possesses a valid copy of the content, which is expected to remain valid for an extended period.

In scenarios where the names match, but the freshness period is transient (indicating a short period), the algorithm takes a different course of action. Instead of caching the existing content, it fetches the new updated copy of content ck directly from the producer. Subsequently, it forwards this updated content to the intended interface.

When the names of the caching node and the potential cacher do not match, and the freshness period is either transient or non-transient, the algorithm decides to forward the DATA packet to the next station. This circumstance suggests that the content is not intended for the current caching node but is designated for another destination within the network.

If none of these specific conditions are met, the algorithm concludes by dropping the DATA packet. It implies that the DATA packet does not align with the criteria for caching or forwarding and is not considered further within the processing pipeline. These nuanced decisions are fundamental in managing the storage and dissemination of content within the context of named data networking.

In summary, Fig. 5 effectively manages the caching and forwarding of DATA packets based on specific conditions related to content freshness, producer-consumer relationships, and the local residual time, ensuring optimal utilization of network resources.

Cache hit ratio

Emphasizing the cache hit ratio as the primary metric for evaluating popularity placement optimization is essential because it is a key indicator of system efficiency. A heightened cache hit rate reflects efficient cache utilization, indicating that a significant portion of requested content is readily accessible within the network. This optimization aligns seamlessly with the overarching objective of improving system performance and enhancing user experience. By strategically siting popular content closer to users, the cache hit rate is maximized, resulting in minimized retrieval delays and reduced network congestion. The article employs Eq. (6) to evaluate the cache hit ratio, offering a comprehensive and quantitative measure of the efficacy of cache placement strategies. This performance metric facilitates a thorough assessment, considering factors such as content availability, network responsiveness, and the overall optimization of content delivery in content-centric networks.

(6) $$CHr=\frac{{\sum }_{i=1}^N(numi)}{Requestedcontent}$$

$V=Setofnodesinthenetworktopology.$

$Numi=Numberofhitsatnodei.$

Impact of hit ratio based on cache capacity

The cache size significantly influences the cache hit rate. Generally, as the size of the cache increases, the likelihood of storing frequently accessed or popular items also rises. This, in turn, results in an increased cache hit rate. A larger cache enables a more extensive retention of frequently requested content, diminishing the necessity to fetch data from the underlying storage or network. Consequently, this improvement enhances the overall efficiency of the caching system.

According to Fig. 6, performance metrics vary as the cache size of the edge domain ranges from 0.2% to 0.5%. As anticipated, increased cache size enhances performance across all considered schemes. Larger storage capacity enables more content storage, thereby reducing delivery time and hop count, leading to advantages in NDN traffic exchange.

Figure 6 illustrates that the DPCP solution exhibits the lowest hit ratio compared to other schemes. It is attributed to the DPCP solution’s focus on calculating content popularity using the request count, which may not effectively improve the hit rate given dynamically changing popularity trends for specific content items. On the other hand, the PaCC caching scheme performs slightly better than DPCP. When distributing popular content among edge node caches, it considers the requested count and popularity at the previous interval. Nevertheless, it fails to capture the history of requesting content items, increasing the cache miss ratio.

In contrast, PaCPn determines the content popularity based on lagged ‘Popularity-score’ and ‘Request-count’ values. The proposed controller-aware VAR popularity model is trained concerning these factors to anticipate future popularity.

Additionally, when the requested content is not cached in the network, it first calculates its local-residual freshness period when receiving it from the producer to limit the miss ratio while contributing to an improved cache-hit ratio. In Fig. 6, ‘DPCP’ achieves a cache hit rate of 63%, while ‘PaCC’ reaches 75% with an expanding edge domain cache. ‘PaCPn’ stands out by achieving a 90% hit rate through dynamic updates of content popularity using a VAR model on an SDN controller.

Impact of hit ratio based on Zipf distribution

We observe the impact of the Zipf distribution parameter $\alpha$ on the cache hit ratio. As $\alpha$ increases, user requests tend to concentrate on the most popular content. This skewed distribution, associated with higher $\alpha$ values, indicates that a small subset of items is frequently requested, creating a long tail’ distribution. Consequently, during a sudden surge in requests for popular content, the cache hit rate tends to rise. This behavior aligns with the Zipf distribution’s representation of skewed popularity, emphasizing the significance of comprehending and adjusting $\alpha$ when optimizing caching strategies to capture and serve the most requested content efficiently.

In Fig. 7, the ‘DPCP’ solution shows the lowest hit rate compared to the others. In contrast, the ‘PaCC’ solution performs slightly better than ’DPCP,’ thanks to the dual-factor-based content popularity prediction used in calculating content popularity. Conversely, the proposed solution, ‘PaCPn,’ achieves the highest hit ratio compared to the others. It accomplishes this by selecting popular content based on its historical record, effectively capturing user attraction towards them. Under the considered settings in Fig. 7, ‘DPCP’ achieves a hit rate of 54%, ‘PaCC’ reaches 68% when the Zipf parameter $\alpha$ increases, but ‘PaCPn’ excels at 80% by considering changing consumer behavior for improved predictions.

Impact of hit ratio based on exchanged CR traffic

Network traffic plays a pivotal role in shaping cache hit rates in an NDN environment, where content retrieval is based on names rather than locations. Efficient and streamlined data transfer in optimal traffic conditions increases cache hit rates. When network traffic is well-managed, the popular selection of content (driven by the multivariate VAR model) is readily available in local caches, minimizing latency associated with fetching data from distant sources while enhancing the cache hitting rate.

In the event of suboptimal traffic conditions, such as congestion or inefficient data transfer, there is the potential for a decrease in cache hit rates. It underscores the need for the programmable nature of the NDN environment, which becomes particularly crucial in addressing challenges associated with suboptimal traffic conditions. The proposed solution leverages programmable mechanisms, effectively mitigating congestion and optimizing data transfer. It, in turn, ensures a more robust and responsive caching system even under less-than-ideal traffic conditions.

As a result, it is evident that ‘PaCPn’ excels in comparison to ‘DPCP’ and ‘PaCC’ by considering ‘Request-count’ and ‘Popularity score’ over pre-intervals to populate cache nodes with popular content, increasing the hit ratio and reducing server load. This approach significantly boosts the cache hit ratio, substantially reducing the amount of NDN traffic within the network. As more user requests are efficiently served from the cache due to the higher hit ratio, there is less dependency on external sources, resulting in a reduced exchange of traffic within the NDN network. Figure 8 shows DPCP achieving a 55% hit rate, PaCC at 65%, and PaCPn surpassing both with an 80% hit ratio. This superior performance is attributed to considering previous lag values and enhancing new content popularity prediction.

Content retrieval delay

Content retrieval delays are critical for evaluating optimal placement in the programmable NDN caching environment. Based on the time taken to retrieve requested content from the cache, these delays significantly impact the efficiency of cache placement. The strategy involves storing frequently requested content in proximity-aware caches, reducing the need for extensive data transmission. It not only enhances the user experience but also optimizes network resource utilization. In the programmable NDN caching landscape, focusing on reducing retrieval delays is key to overall performance, ensuring optimal content delivery responsiveness. The retrieval delays can be computed using the formula defined in Eq. (7).

(7) $$RD=\frac{{\sum }_{i=1}(RDi)}{\frac{|re{q}^c|}{|consumers|}}$$

$RDi=Retrievaldelayofasinglecontenti.$

$re{q}^c=Totalamountofrequestedpackets.$

$sub=Totalnumberofcontentconsumers.$

Impact on retrieval delays based on cache capacity

The influence of cache size on content retrieval delays is a significant factor in shaping the efficiency of a content delivery system. In essence, the size of the cache directly impacts the system’s ability to store and quickly retrieve frequently requested content. A larger cache facilitates the storage of popular items, reducing the need to fetch data from underlying storage or network sources. Consequently, with a generous cache size, the content retrieval process becomes faster and more responsive as the likelihood of finding requested data in the cache increases. On the contrary, a smaller cache may lead to more frequent instances of content not being readily available, resulting in longer retrieval delays as the system resorts to fetching data from external sources. It underscores the critical role of optimizing cache size for minimizing content retrieval delays and enhancing the overall efficiency of a content delivery system.

In Fig. 9, DPCP shows the highest retrieval delays as the cache size increases, primarily due to its dependence on content popularity score, which may not effectively reduce delays with larger caches. PaCC performs slightly better by considering popularity and distance but is compromised by a hop count-based node selection approach, lacking consideration for factors like link cost and latency, leading to inefficiencies. Multiple updates in PaCC for optimal content placement make it computationally expensive, especially with larger caches.

In contrast, PaCPn employs a centralized caching policy managed by the SDN controller to minimize computational costs and optimize caching decisions, resulting in superior performance with larger cache sizes. The PaCPn caching algorithm makes placement decisions based on distance cost, delivery time, and popularity status to reduce delivery times and network delays. It uses a closeness-aware metric considering ‘content delivery time’ and ‘content distance cost.’ Delivery cost reflects the link cost, and distance cost measures the caching node’s proximity to the user. The strategy also accounts for content popularity, optimizing placement for frequently requested content.

In comparative performance evaluations, the dynamic popular content placement (DPCP) strategy achieves a retrieval delay of 0.07 ms, showcasing its efficiency. The popularity-aware caching (PaCC) strategy attains even lower delays of 0.06 ms, which is particularly noteworthy with an expanding edge cache size. However, the PaCPn algorithm stands out, surpassing all metrics by demonstrating superior performance with delays reduced to an impressive 0.046 ms.

Impact on retrieval delays based on number of consumers

The impact of the number of consumers on content retrieval delays is a crucial factor in determining the responsiveness of a content delivery system. As the number of consumers increases, the system experiences a higher demand for content, potentially leading to increased retrieval delays. A more extensive consumer base results in more frequent requests for various content, placing additional demands on the system’s resources. In scenarios where the optimized placement strategy is not adequately scaled to accommodate the growing number of consumers, retrieval delays may occur as the system struggles to meet the heightened demand. Conversely, retrieval delays can be minimized in an environment with a well-scaled infrastructure capable of handling a more extensive consumer base. Therefore, optimizing the placement strategy to align with the number of consumers is crucial for ensuring efficient content retrieval and maintaining satisfactory performance even under increased demand.

As illustrated in Fig. 10, all the considered schemes exhibit lower retrieval delays as the number of consumers increases. Especially notable is the proposed solution, which demonstrates the lowest delay compared to other schemes when responding to DATA requests from consumers. It underscores the potential for optimizing content retrieval delays by intelligently managing caching decisions and placement, especially in scenarios with increased consumers and concentrated requests for popular content.

The solution ensures that even with the growing number of consumers, the most popular content is cached at the closeness edge points, reducing retrieval delays and enhancing overall content delivery efficiency. The experimental results reveal distinct performance variations among caching policies. DPCP exhibited a retrieval delay of 0.09 ms, PaCC demonstrated efficiency with a content delivery delay of 0.06 ms, and notably, our proposed caching scheme outperformed both, achieving an impressive delivery rate of 0.04 ms. This superior performance is attributed to our innovative heuristic algorithm, dynamically allocating caching resources to popular content. The dynamic allocation ensures strategic placement, reducing retrieval delays and enhancing the content delivery experience.

Impact on retrieval delays based on network throughput

The influence of system throughput on content retrieval delays is of crucial significance, directly impacting the speed and efficiency of delivering requested content. System throughput, representing the number of data packets processed within a given timeframe, is crucial in determining how swiftly content can be retrieved and transmitted to consumers. Higher system throughput correlates with faster content delivery, minimizing retrieval delays and ensuring a more responsive user experience. In contrast, lower system throughput can lead to increased retrieval delays as the system struggles to process and transmit content efficiently. Therefore, optimizing system throughput is instrumental in achieving prompt and reliable content retrieval, directly contributing to users’ overall performance and satisfaction interacting with the content delivery system. As a result, the delivery time is reduced because devices are not individually making cache placements and forwarding decisions; the controller dynamically implements closeness-aware placement decisions at regular intervals, unlike the other schemes in which distributed placement decisions are performed through the betweenness centrality approach. This approach allows a larger content volume to be efficiently delivered within shorter time frames, enhancing the overall network throughput of DATA packets in bits per second (DPBS). Figure 11 shows that the DPCP solution reaches a minimum delay of 0.07 ms. The other scheme achieves a minimum retrieval delay of 0.05 ms as the system throughput increases. On the other hand, the proposed solution PaCPn achieves the lowest content delivery rate of 0.033 ms due to its centralized and dynamically managed placement of popular content in popular NDN devices.