technologies such as afordable processors, fast internet access, and standardization. This development has been With the introduction of packet-switched networks, it driven by the need to bridge the gap between high user became possible to distribute multimedia content that demands and limited network and storage capacities. For combines data with audio and video information. In example, a digital video signal of "television quality" refact, once digitized, audio and video information is trans- quires 216 Mbit of storage transmission capacity for one ferred in byte format. The initial dificulties encountered, second of video, while a movie with a duration of about however, involved the need for increasingly performant two hours requires over 194 GB of space, values that exconnections between network nodes in terms of band- ceed the current capacities of networks. Video CODECs width and throughput for real-time applications such are essential as they allow data compression, facilitating as internet telephony (VoIP) and streaming, the latter transmission and storage. having become predominant in internet trafic. Most video CODECs currently in use conform to one The main transport protocol for these applications is of the international standards for video encoding. Among the Real-Time Protocol (RTP), which operates over lower- the most influential, JPEG has become the standard for level protocols like UDP [1, 2, 3]. With the constant storing still images, while MPEG-2 and its evolutions, increase in IP trafic, a significant rise in video trafic such as MPEG-4, are essential for digital television and is expected, which will constitute a large part of global DVDs [9, 10, 11, 12, 13, 14, 15, 16, 17]. Other video internet trafic. CODECs on the market include Divx, Xvid, VP8, VP9, The growth of the sector involves various players, from and VP10. Telcos, telecommunications operators overseeing technological infrastructures, to "Over The Top" (OTT) entities, 2.2. Cloud Computing which come from the digital services and technology sector and need an infrastructure to provide their services The term Internet of Things (IoT), coined by Kevin Ash[4, 5] and energy systms [6, 7, 8]. The transition from a ton, refers to a set of devices connected to the Internet, "voice-centric" to a "data-centric" communication model such as mobile phones, cofee machines, and many other has required new interconnection models and services objects [18, 19, 20, 21]. such as Content Delivery Networks (CDNs) to improve In 2003, there were about 6.3 billion people living on the Quality of Experience (QoE) for users. the planet and 500 million devices connected to the InThis work focuses on improving QoE through Trans- ternet, well below the threshold to consider the IoT as parent Caching platforms, analyzing technologies and such. mathematical models, with the aim of optimizing access The explosive growth of smartphones, tablets, and PCs to multimedia content on modern networks such as LTE. brought the number of devices connected to the InterThe work is structured into various chapters that address net to 12.5 billion in 2010, while the world population technical and practical aspects related to these topics. increased to 6.8 billion. It is precisely from this moment that the expansion of the IoT, according to Cisco IBSG's definition, begins [22]. 2. Services The eficient and stable use of IoT technologies is linked to the data management and processing capa2.1. Codec Video bilities, significantly improved with cloud computing platforms such as fog computing or edge computing. The main advantages of using cloud computing are:
In recent years, image and video encoding has become crucial for many digital applications, thanks to advanced • cost reduction; • faster access;; • economies of scale; • improvement of overall performance; • greater security.
3.1. Content Distribution: "The The function of a Transparent Caching platform is similar to that of traditional CDNs, which is to ensure that
Bottleneck Theory" content is placed near the end users, thereby reducing The concept of content distribution arises from the sim- both the round trip time (RTT) and packet loss, thus ple necessity of having content closer to the customers improving throughput performance and Quality of Exwho request it. Content distribution technologies have perience. The main diference from traditional CDNs become pioneers in addressing some of the issues gener- is that Transparent Caching platforms use self-learning ated by the rapid growth of the Web, such as slow content functionalities, meaning they dynamically learn the most download times and WAN link congestion. important content (generally the most requested) and
The objective of this work is to measure the quality then classify them in order of importance, i.e., based on of certain connections to solve the problem of the "dis- the most viewed. The content is then stored on a platform tant" location of some servers, which should instead be and made available only on demand. In addition to the "replicated" near the client (requester) to optimize the traditional functions of a CDN, Transparent Caching platuser experience and resource utilization. forms are often used for IP trafic that carries content not
A data flow can traverse a network only at the speed considered stable (such as web content) or for trafic for allowed by the slowest link in its path. In theory, every which there is no agreement with the Content Provider. route through a network has a potential "bottleneck. The The use of Transparent Caching platforms ofers multiple necessity to improve the throughput is crucial for Non- advantages: Terrestrial Networks (NTNs) [23]. • no client configuration required, reducing admin
These critical points, often located at the "edge" of the istrative activities; Internet, are primarily influenced by bandwidth. In the past, corporate networks followed the 80/20 rule (80% • increased application throughput; of trafic on the LAN and 20% on the WAN), but with • decreased necessary bandwidth; the expansion of the Internet, this balance has reversed, • greater reliability. leading to an overload on the WAN. Companies, due to In summary, Transparent Caching platforms ofer a flexieconomic constraints, cannot increase the bandwidth of ble and efective solution for improving network perforWAN links, making these critical points the main sources mance, optimizing content access for end users. of congestion.
Content Delivery Network or Content Distribution Net- 4.1. Definition work is represented by a group of network servers located in specific data centers around the world with the Network performance is evaluated through the measurepurpose of facilitating the faster retrieval of content for ment of appropriate indicators, of which throughput Internet users. These servers cache all the information (data flow) and latency are the main ones. It is then requested by a content provider. useful, if not essential, to consider the quantities derived
CDNs improve the distribution of various types of con- from these. tent, including high-definition videos, audio streams, and software downloads. The use of CDNs ofers significant 4.1.1. Bandwidth advantages such as:
Bandwidth is the capacity of the physical channel avail• improvement of Quality of Experience and down- able to transfer a certain amount of information (in bits) load times; relative to the considered time interval. • protection of servers from overloads and inter
ruptions; 4.1.2. Throughput (TH or THR) • flexibility in managing static content; • ease in handling access spikes and increased security against attacks.
Throughput refers to the amount of data trafic and information that actually reaches its destination within a unit of time, net of network losses and protocol operations.
Thus, throughput is linked to the Quality of Experience perceived by the end user.
In summary, CDNs represent a fundamental infrastructure for the eficient distribution of content on the Internet, making the user experience smoother and more secure. 4.1.3. Latency Latency is the time required for a message to reach the destination node and is measured in units of time, usually seconds.
In this work, latency is considered as a performance indicator, defined as the time required for a message to travel from a source node to a destination node and then back to the source node. This time interval is called the Round Trip Time (RTT) of the network.
Packet loss is a good measure of the quality of the connection (in terms of packet loss rate) for many TCP-based applications. It is generally caused by congestion, which in turn causes queuing of messages waiting to be transferred (e.g., in routers).
A study conducted by Cisco on global Internet trafic forecasts and trends for the period 2017-2022 has shown that mobile data trafic during this period will increase exponentially, reaching approximately 77.5 Exabytes per 4.1.6. Bit Rate (BR) month by 2022 [24]. This volume will represent 20% The bit rate (or transmission rate) is the number of bits of total IP trafic. To cope with this sudden increase in per second; that is, the amount of digital information data trafic, new technologies have been introduced to (bits) transferred per unit of time (second). Generally, it enhance data transmission speeds and make frequency determines the size and quality of video and audio files: spectrum usage more eficient. Additionally, new frethe higher the bit rate, the better the quality of the video quencies for mobile radiocommunication have been inand the larger the file size. troduced through an increase in the number of radio cells. This new technological approach has resulted in an increase in spectrum eficiency of about three times that of 4.2. Mathis Model its predecessors, with significantly lower network costs. The Mathis model explains the relationship between LTE stands for Long Term Evolution; the telecommunithroughput and packet loss through the following for- cations body known as the Third Generation Partnership mula: Project (3GPP) initiated the project in 2004, although this = new technology first entered the market starting in 2010.
* √ LTE represents a broadband wireless technology dewhere: signed within the telecommunications market. According to 3GPP, a series of advanced requirements have been identified for using this new technology: • MSS (Maximum Segment Size) is the maximum
segment size; • TH is the throughput; • RTT is the round trip time; • p is the probability of packet loss; • C is a constant with a value equal to √︁ 23 .
For simplicity, the values of p and the maximum segment size (MSS = 1460 bytes) are assumed to be constant.
The Mathis formula, therefore, highlights how the value of throughput is inversely proportional to the probability of packet loss and latency, but directly proportional to the segment length, as summarized in the following figure. • Reduction of cost per bit • Increased service provisioning: more services at
lower costs with improved user experience • Flexibility in using existing and new frequency
bands • Simplified architecture, open interfaces • Allowing reasonable terminal power consump
tion
The primary goal of LTE is to provide high data transmission speeds and increase capacity, improve coverage, achieve low latency, and optimize packet access technology to support flexible bandwidth implementation.
At the same time, its network architecture has been designed to support packet-switched trafic, seamlessly and with high quality of service.
The LTE standard, therefore, supports only packetswitched communication with its all-IP network. The reason LTE is designed exclusively for packet switching is that it aims to provide uninterrupted IP connectivity between user equipment and the packet data network, without interrupting end-user applications during mobil- Figure 2: Scenario 1 ity [25].
The experimental activity presented describes the use of a drone equipped with a camera payload and a link to a ground station for transmitting a video made in accordance with the assigned mission, in 4K resolution, followed by uploading it to a transparent caching system.
The drone’s function is simulated; the drone sends video to the server, which processes the received data. The server then sends the data to the client upon request.
Specifically, this work focuses on the activity described, for which a campaign was conducted to measure the quality of the link established between a client located at the University of Tor Vergata or at Villa Mondragone in Frascati, who wishes to view or download the video from a server located in Paris, Île-de-France.
The measured data will lead to a series of results representing a significant sample for evaluating the benefits derived from using a transparent caching function and estimating how this position afects final performance. In particular, simulations of transparent caching located at progressively decreasing distances from the client will be used in two diferent scenarios (e.g., in the first scenario, the client is located at Villa Mondragone, the server at Îlede-France, and the transparent caching servers in Milan, Rome, and Tor Vergata).
In terms of latency, it is expected that approaching the server will result in a reduction of this metric, leading to an increase in throughput.
6.3.1. Ping The ping function allows measuring the response time (round-trip time in milliseconds (ms)), packet loss percentages, variability in response time both in the short term (seconds scale) and long term, and the lack of reachability, i.e., no response for a series of pings.
Network processing eficiency largely depends on the network’s ability to ensure an adequate level of performance, which must be measurable. Specifically, as previously illustrated, throughput is preferred, as it reflects actual measured performance rather than the maximum bandwidth available on the line. The second measurable value characterizing performance is latency, which corresponds to the time required for a message to traverse the network from the source node to the destination node and back to the source node. It is measured exclusively in terms of time.
A single physical line connecting the same two computers continuously has a constant RTT value, while TCP connections are likely to exhibit very diferent RTT values. For example, a TCP connection between two cities thousands of kilometers apart might have an RTT of 100 ms, whereas a TCP connection between two computers in the same room, only a few meters apart, might have an RTT of 1 ms, and the same TCP protocol must accommodate both connections. Additionally, to complete the scenario, the connection between the two cities might vary significantly within the same day, and even variations in RTT values are possible during a TCP connection lasting only a few minutes. 6.4. Tools For conducting the experimental campaign, the following tools, data processing devices, and network equipment were used: • iperf3; • smartphone with Android operating system; • PC with Windows 10 operating system.
The experimental activity was based on measuring the RTT (Round Trip Time) using the PING functionality for decreasing distances, following the schema represented below and assuming that the service requester (client) was always located at the Villa Mondragone site.
The distance between the fixed client station and the servers simulating the transparent caching functionality at increasing distances also serves to hypothesize the theoretical value of RTT, which can be confirmed by the ping measurement. Some measurement samples confirm the validity of this reasoning.
In the first scenario, Villa Mondragone was considered the access site (client) from which requests were made to the server at decreasing distances. In the initial phase, measurements were conducted using the IPERF3 tool, and these measurements were found to be consistent with those of the subsequent phase. A PING request was then sent to simulate access to the server located in Île-de-France, approximately 1450 km away. RTT measurements were taken with the PC (Villa MondragoneÎle-de-France) for diferent packet lengths, expressed in bytes. Subsequently, using the Mathis law, the throughput values for various PLR (0.1% and 0.37%) were derived, as shown in the following figure:
As in the previous case, this simulation considers the client located at Villa Mondragone, while the server being queried is positioned in Milan. In this scenario, it is as if a "transparent caching server" were used, placed in locations closer to the client compared to the server where the videos are stored. For this connection, RTT measurements were taken with the PC for the distance of approximately 600 km (Villa Mondragone-Milan) with various packet sizes, expressed in bytes. Subsequently, applying the Mathis formula, the throughput values were derived for diferent Packet Loss Rates (PLR) of 0.1% and 0.37%, as shown in the following figure:
Similar to Case 2, in this simulation, the client is at Villa Mondragone, while the server is located in Rome. For this connection, RTT was measured with the PC for the distance of approximately 30 km (Villa Mondragone-Rome) with diferent packet sizes, expressed in bytes. Using the Mathis formula, the throughput values were derived for various PLR values of 0.1% and 0.37%, as illustrated in the following figure:
In this case, as in the previous ones, the client is at Villa Mondragone, and the server is positioned at the University of Tor Vergata, Department of Information Engineering. RTT measurements were taken with the PC for the distance of approximately 10 km (Villa Mondragone-Tor Vergata) with various packet sizes, expressed in bytes. The Mathis formula was then used to derive the throughput values for diferent PLR values of 0.1% and 0.37%, as shown in the following figure:
In this final case, both the client and server are located at Villa Mondragone. For this connection, RTT was measured with the PC for the distance of approximately 0 km (Villa Mondragone-Villa Mondragone) with various packet sizes, expressed in bytes. The throughput values were then derived using the Mathis formula for diferent PLR values of 0.1% and 0.37%, as depicted in the following ifgure:
The experimental activity was based on measuring RTT using the PING functionality for decreasing distances, following the schema represented below, assuming that the service requester (client) was always located at the University of Tor Vergata, Department of Information Engineering.
The distance between the fixed client station and the servers, which simulate the behavior of transparent caching at increasing distances, also helps to hypothesize the theoretical value of RTT, which can be confirmed by the ping measurements. Some sample measurements confirm the validity of this reasoning. In the first scenario, the University of Tor Vergata (Department of Information Engineering) was considered the access site (client) from which to request data from the server at decreasing distances. Initially, measurements were conducted using the IPERF3 tool, and results were consistent with those obtained in the subsequent phase. A PING request was then sent simulating access to the server located in Ile-de-France, approximately 1440 km away. For this connection, RTT measurements were taken with the PC (University of Tor Vergata-Ile de France) with various packet sizes, expressed in bytes. Subsequently, applying the Mathis formula, the throughput value was derived for diferent PLR values (0.1% and 0.37%), as shown in the following figure:
As in the previous case, in this simulation, the client is located at the University of Tor Vergata (Department of Information Engineering), while the queried server is positioned in Milan. In this scenario, it is as if a "transparent caching server" were used, placed in locations closer to the client compared to the server where the videos are stored. For this connection, RTT was measured with the PC for a distance of approximately 595 km (University of Tor Vergata-Milan) with various packet sizes, expressed in bytes. Applying the Mathis formula, the throughput value was derived for diferent PLR values (0.1% and 0.37%), as illustrated in the following figure:
In this third case, the client remains at the University of Tor Vergata (Department of Information Engineering), while the server is located in Rome. For this connection, RTT was measured with the PC for the distance of approximately 13 km (University of Tor Vergata-Rome) with various packet sizes, expressed in bytes. Applying the Mathis formula, the throughput value was derived for diferent PLR values (0.1% and 0.37%), as shown in the following figure:
In this final case, both the client and server are located at the University of Tor Vergata (Department of Information Engineering). For this connection, RTT was measured with the PC for the distance of approximately 0 km (University of Tor Vergata-University of Tor Vergata) with various packet sizes, expressed in bytes. Applying the Mathis formula, the throughput value was derived for diferent PLR values (0.1% and 0.37%), as depicted in the following figure:
From a theoretical standpoint, applying the Mathis law and experimental laws linking distance to RTT yields results consistent with those measured. Specifically, the following aggregated values reflect a decreasing trend of THR as a function of RTT. 7.1.1. PLR=0,1% Tor Vergata (Case 4), and Villa Mondragone (Case 5); • 86 < < 99: THR for Case 1 is above Cases 2 and 4, so the Transparent Caching locations that still maintain eficient performance are Rome (Case 3) and Villa Mondragone (Case 5); • 99 < < 105: THR for Case 1 is above Cases 2, 3, and 4, meaning that the only Transparent Caching location still providing eficiency is Villa Mondragone (Case 5); • 105 < < 127: THR for Case 1 is above Cases 3 and 4. Therefore, the Transparent Caching locations that still ensure eficiency are Milan (Case 2) and Villa Mondragone (Case 5); • > 127: THR for Case 1 is above all other cases, indicating that none of the other Transparent Caching locations ensure eficient performance.
In summary, the final situation of Scenario 1 with PLR = 0.1% is represented in the following figure:
Comparing the various curves for the diferent cases in Scenario 1, with PLR = 0.1%, the following observations are made: • < 82: THR decreases as RTT increases.
In particular, the curve representing the simulation without Transparent Caching (Case 1) lies below all other curves, indicating that the result aligns well with expectations; • 82 < < 86: THR for the case without Transparent Caching (Case 1) is above the curve representing Milan (Case 2). In this interval, the Transparent Caching locations that still ensure eficient performance are Rome (Case 3),
In this case, the behavior of the functions is identical to the case with PLR = 0.1%, while at the same RTT, the THR value is approximately 90% higher than in the previous case. Therefore, it is observed that this second case validates the results already obtained. ensures eficient performance is only Tor Vergata (case 4), which is the same site where the client is located; • > 180: The ECC no longer has any efect, and it is better for the client to query the main server directly.
In summary, the final situation for Scenario 2 with PLR = 0.1% is represented in the following figure:
From a theoretical perspective, applying the Mathis equation and experimental laws linking distance to RTT values leads to results similar to those measured. In particu- 7.2.2. PLR=0,37% lar, the following aggregated values reflect a decreasing trend of THR as a function of RTT. 7.2.1. PLR=0,1% With similar considerations applied for a diferent PLR (=0.37%), the following observations are made:
From the comparison of the various curves for the diferent cases in Scenario 2 with a PLR of 0.1%, the following observations are made: • < 167: THR decreases as RTT increases, and in particular, the curve representing the simulation without ECC (case 1) is below all others, indicating that the result is largely as expected; • 167 < < 172: THR for the case without ECC (case 1) is above that for Milan (case 2), so in this range, the ECCs that ensure eficient performance are Rome (case 3) and Tor Vergata (case 4); • 172 < < 180: THR for case 1 is above that for cases 2 and 3, so the ECC that still
The behavior of the functions is identical to the case with PLR = 0.1%, while at the same RTT, the THR value is approximately 90% higher than in the previous case.
Therefore, this second case also validates the results previously obtained. In summary, the behavior between the two scenarios is consistent. Indeed, for lower extreme values of RTT within the defined ranges for both scenarios, a server without ECC does not appear advantageous, whereas all other sites simulating the ECC function (Milan, Rome, Tor Vergata, and Villa Mondragone) are beneficial. Conversely, for higher extreme values of RTT, the site can be directly accessed by the client, as the values observed on other servers do not justify the use of ECC. In intermediate positions, however, as RTT increases, it becomes increasingly advantageous to use the server closest to the client.
Therefore, in conclusion, the experimental activity measuring RTT in diferent cases confirms and validates the results from the simulation activity based on the Mathis equation.
Considering the percentage variations in RTT across diferent cases diferentiated by server location, the results shown in the following figure are obtained. It is observed that the RTT variation is around 20%, except for the Tor Vergata site, where the variation is approximately 150%. The significant diference noted is certainly attributable to the quality of the LTE connection during the measurements. Given that the average value of the measurements is notably lower and consistent with theoretical predictions, it can be considered that the "anomalous" value might be regarded as a statistically insignificant outlier.