An Enhanced Traffic Split Routing Heuristic for Layer 2 and Layer 1 Services

Virtual Private Networks (VPNs) have now taken an important place in computer and communication networks. A virtual private network is the extension of a private network that encompasses links through shared or public networks, such as the Internet. A VPN is a transmission network service for businesses with two or more remote locations. It offers a range of access speeds and options depending on the needs of each site. This service supports voice, data and video and is fully managed by the service provider, including routing equipment installed at the customer’s premises. According to its characteristics, VPN has widely deployed on ”COVID-19” offering extensive services to connect roaming employees to their corporate networks and have access to all the company information and applications. Hence, VPN focuses on two important issues such as security and Quality-of-Service. This latter has a direct relationship with network performance such as delay, bandwidth, throughput, and jitter. Traditionally, Internet Service Providers (ISPs) accommodate static point-to-point resource demand, named, Layer 1 VPN (L1VPN). The primary disadvantage of L1VPN is that the data plane connectivity does not guarantee control plane connectivity. Layer 2 VPN is designed to provide end-to-end layer 2 connection by transporting layer 2 frames between distributed sites. An L2VPN is suitable for supporting heterogeneous higher-level protocols. In this paper we propose an enhanced routing protocol based on Traffic Split Routing (TSR) and Shortest Path Routing (SPR) algorithms. Simulation results show that our proposed scheme outperforms the Shortest Path Routing (SPR) in term of network resources. Indeed, 72% of network links are used by the Enhanced Traffic Split Routing compared to Shortest Path Routing (SPR) which only used 44% of the network links. Keywords—Virtual private network; enhanced traffic split routing; quality of service; shortest path routing; layer 1 VPN; layer 2 VPN


I. INTRODUCTION
Computing, and networks in particular, have changed a lot over the past twenty years. The flow of information and the emergence of new technologies have increased considerably [1]. It is now possible to exchange substantial data of all types as well as to transmit voice and video over computer networks [2].
Indeed, with a modern economy based on new information and communication technologies, most companies use a set of means for the implementation of a reliable and flexible computer network [3]. This network allows corporate users to share resources such as printers, files and data. As a result, the need for remote connection to corporate resources has become common. Remote applications thus become the main tool of the company's information system. The question that may arise then is how to ensure access within a structure sometimes spread over large geographical distances? In concrete terms, how can a branch of a company access data located on a server in the headquarters several thousand kilometers away?
Virtual private networks (VPNs) have been set up to respond to this type of problem and takes an important place in computer and communication networks. As pointed out by [4], a VPN is the extension of a private network that encompasses links through shared or public networks, such as the Internet.
It offers a range of access speeds and options depending on the needs of each site. This service supports voice, data and video and is fully managed by the service provider, including routing equipment installed at the customer's premises [5]. Indeed, the damaged caused by "COVID-19" on global economy leads company networks on looking for VPN solutions to establish a private communication to the corporate intranet while traveling from home.
Both service providers and customers are starting to realize the benefits of VPN solutions. New applications such as voice, telemedicine and video on demand make it possible to envisage an increase in productivity and a reduction in costs. However, VPNs are not only interested in extending LANs at a lower cost, but also in the use of specific services or functions ensuring quality of service (QoS) and security of exchanges [6] [7]. Indeed, the notion of quality of service makes it possible to formalize the requirements for each type of service in terms of performance criteria: bandwidth, end-toend transmission delay, packet loss rate, jitter, etc. Each service may have different quality requirements. Services such as voice or video impose very strong constraints on the quality of transmission: transmission delays or data loss must not degrade communication or the broadcasting of a video stream. In order to support real-time and multimedia applications on virtual private networks, it is necessary to develop routing algorithms which take QoS parameters into account. Routing algorithms with QoS must be adaptive and flexible for efficient management of resources in the network [8]. In practice, routing with QoS has not worked well.
The objective of routing is to determine a route (i.e. a set of links to be traversed), respecting certain constraints, to establish a connection from a source node to a destination node. The purpose of a routing algorithm is to allow the calculation of the route between those two nodes within the meaning of a certain criterion?
The remainder of the paper is organized as follows. The next section will focused on different algorithms such as Waxman and Brite algorithms. Section 3 highlights the approach proposed in this work. Then, an analysis of the performance of our prototype system implemented using simulation model will be described in Section 4. In Section 5, numerical results are presented to show the effectiveness of the proposed algorithm. Lastly, the conclusion and future work are outlined in Section 6.

II. LITERATURE REVIEW
VPNs allow remote users, partners and providers to access certain parts of their networks (intranets). They also allow the deployment of many types of applications such as realtime voice or video, critical business management software or interactive applications [9] [10]. Originally, companies using VPN solutions used "layer 1" services such as "leased lines" (and referred to as layer 1 VPN, L1VPN). Leased lines are dedicated connections that a telecom operator operates directly between two customer sites, providing a permanent connection at a determined speed. Although leased lines offer users the confidentiality and reliability of transferred data, they suffer from a lack of flexibility compared to other types of layer 2 solutions such as Frame Relay, ATM, L2TP, L2F or also more recently Carrier/Metro Ethernet.
However, as mentioned in [11] [12], this type of VPN is characterized by its prolificity in singular domain and the lack the Quality of Service (QoS) during the inter-domain routing which lead to inhibit its scalability and flexibility. Another issue with L1VPN is the inter-configuration of a customer on another Service provider network as the policies are distinctives. This can be solved using the address mapping mechanism, unfortunately , this latter is not well-defined in standard specifications [13].
In fact, "layer 2" VPN services (or also layer 2 VPN, L2VPN) have allowed service providers to offer their customers a connection similar to that offered by leased lines. On the other hand, with L2VPN it is no longer necessary to have a dedicated leased line for each network interconnection. The clients share a single physical line and each has its own logical channels to send its traffic [14]. Layer 2 VPN services are attractive to the service operator because they do not require the operator to participate in the design and configuration of layers 2/3 of the customers' LAN. Also, the management and maintenance of the control plan are carried out by the customer and they are transparent to the operator's network [15] [16].
The major problem with L2VPN is security. Unlike L1VPN where each customer has their own private line, a layer 2 VPN is deployed on a shared network infrastructure that can be managed by national and international network service providers. As a result, several companies disagree that their data should be transferred through shared, unsecured tunnels [17]. One solution would be to offer a layer 3 or L3VPN VPN service. Security is usually provided by a combination of tunneling and encryption methods. The best known is the one that implements the IPSec (IP Security) protocol. IPSec is a "layer 3" security protocol. It is based on the IP protocol and offers tunneling and security features, including encryption, authentication and key management [18]. In this work, a random generator graph named "Brite" is used [19]. The BRITE topology generator assigns each link with a delay based on its physical distance. The algorithm can be described as follows [20]: • Firstly, we specify the number of nodes on the networks.
• For a link creation between two nodes u and v we define the probability P (u, v) : Where, • d(u, v): the distance separating from node u to node v; • L: the maximum distance between node u and node v; • α and β: These two constant parameters are defined in the interval (0.1]. When the constant α is decreased, we noticed that the link's density on the network is increased. Based on the link probability (u, v) a link is added or not between u and v. In a shortest path tree problem, we consider a directed graph G = (V, E), where V represents the set of nodes and E the set of links. Each edge has a weight P i . A path C =< e 1 , e 2 , ..., e n , > has a weight which represents the sum of the weights of the edges constituting the path. The shortest path from a vertex d to a vertex a is the minimum weight path that connects d to a [21].
The two algorithms Bellman-Ford and Dijkstra described in [22] are two well known shortest path algorithms. Shacham [23] proposed a maximum bandwidth tree algorithm to distribute data hierarchically. It uses an algorithm close to Dijkstra to calculate the maximum bandwidth of a single path to all destinations.
The principle of this algorithm is as follows: 1) Determine the maximum bandwidth paths available between the different nodes. 2) Sort the receivers according to their reception capacities. 3) Add recipients to the maximum bandwidth tree one by one.
This hierarchical distribution approach gives for each individual receiver the rate at which it will receive data from the source. The bandwidth will then be allocated appropriately.

III. PROPOSED ALGORITHM
As aforementioned, there are many issues on deploying Dijkstra and Bellman-Ford for routing protocols using two or more constraints. Dijkstra as well Bellman-ford are deployed where we using one constraint and offer good paths. However, when we need the formation of a balanced system when distributing the load this is not guarantee by Dijkstra and Bellman-ford which use an order of priority in the constraint's choice. Our proposed scheme, named Enhanced Traffic Split Routing (E-TSR), is bas on algorithm Traffic Split Routing (TSR) [24] [25] which offer a good objective on load balancing inside a network. Enhanced Traffic Split Routing try to distribute homogeneously the traffic on the network and offer a balanced sharing of traffic. Indeed, with E-TSR, the maximum possible of links are used to balance the traffic on the network.
We present in what follows, algorithm 1, the heuristic of traffic distribution used which is an enhanced algorithm based on [24] [25]. To begin, it is interesting to note that is not always optimal to use the shortest path between a pair of nodes "i" and "j". Accordingly, we will use a model of an M/M/1 queue. Suppose that between two nodes "i" and "j" we have two paths: the shortest path of length "n" and another longer path of length m > n .
To begin with, we assume that we have a first path calculated by the shortest path algorithm. This path links a source "i" to a destination "j" and uses "n" links. We assume that each link in the path from "i" to "j" is modeled by an independent M/M/1 queue (Kleinrock independence assumption) as illustrated on Fig. 1. Now suppose that the traffic is shared between the path with "n" hops and that of "m" hops. Consider the following variables: ρ: use of the link, when all the traffic is offered to the first path only (the shortest path) we have the average residence time on a link: Therefore, the use of the link Since this path is composed of "n" independent links, the average residence time in the path is modeled by the following formula: If we share the traffic between the two paths; the first made up of "n" hops and the second of "m" hops, we have the variables: • λ 1 : the arrival rate for path 1 • λ 2 : the arrival rate for path 2 So, the average residence time across the two paths is: If we find cases where T 1 < T 2 then we can state that the shortest path does not give precisely the best delay. We can say that: Where, In the case where m < n, then the inequality 6 gives us: The inequality 12 shows that the number of hops in the path should be small and not greater than a certain constant. Additionally, this inequality clarifies that when the shorter path is overloaded (maximum link utilization), using another longer path to route traffic can be useful in order to reduce the wait time. Moreover, we can deduce that the number of hops in the longest path decreases when the load offered to this path (ρ 2 ) increases. In particular, in the case where the traffic is distributed between the shortest path and the longest one (ρ 1 = ρ 2 ), it suffices to have m < n/((1 − ρ)) to reduce the waiting time by traffic distribution. For example, if ρ = 80%, then we must have m < 5n. That is, the number of hops in the longest path should not exceed 5 times the number of hops in the shortest path.

IV. RESULT AND DISCUSSION
This section is devoted to evaluating the performance and quality of services resulting from the Enhanced Traffic Split Routing (E-TSR) algorithm by comparing the results with those obtained with Traffic Split Routing (TSR) [25] and the Shortest Path Routing (SPR). In order to be able to evaluate these three algorithms, various simulations were carried out using the NS-2 simulation platform [26].
To begin, we attempt to give an overview of different parameters related to evaluate the performance of our proposed scheme. Then, we will present more closely the NS-2 tool as loop: waiting for a new Virtual Private Network connection demand from any network's node 5: Complete (or Generate) a path (tree) coupling all the new Virtual Private Network and avoiding links whose L s > n 6: if path is icomplete then 7: n ← n + 1 and goto step 5 8: The rest of this section will highlights the QoS parameters used to evaluate the three heuristics: the enhanced traffic distribution (TSR, Traffic Split Routing), the traffic distribution (TSR, Traffic Split Routing), and the shortest path (SPR, Shortest Path Routing). All the simulation parameters are given in Table I. Our simulations are performed using the NS-2 network simulator. For accuracy and compliance, all simulations are performed Twenty times for each scenario. All simulations are performed to study the behavior of the three routing algorithms; E-TSR, TSR, and SPR. All simulations are generated with different random number seeds and the results are averaged over all the outcomes. Fig. 2, Fig. 3, and Fig. 4 illustrate an example of network scenario used in performance study for the three heuristics E-TSR, TSR, and SPR. Fig. 4 illustrates the scenario using Shortest Path traffic algorithm. From this figure, we suppose that the traffic is sent from node 2 to the destination node 4. We see that the node named 6 used as a Steiner node and all traffic is focused on the shortest path (from node 2 to node 6, and from node 6 to node 4). On the other hand, Fig. 2 illustrates the scenario of the Enhanced Traffic Split Routing. From this Figure we can see that the traffic is shared equitably between different paths. Indeed, the traffic sent from node 2 and has as destination node 4 has taking different paths (from node 2 to node 6, from node 2 to node 7, etc.). with this approach, we can see that we use maximum links on the network.   subsequently, with 16 source VPNs, the average throughput with E-TSR is equal to 5.35 Mbps while it is equal to 3.93 Mbps with SPR algorithm. We see also that the gap on the average data rate increase with the number of source VPNs and the E-TSR algorithm offer good throughput compared to TSR and SPR algorithms. This is due to algorithm properties. Indeed, with SPR, all the traffic is focused on the shortest path while with E-TSR the traffic is divided between the maximum number of links on the network. Moreover, the use of maximum links allowed networks to offer a higher throughput especially for networks with large traffic, unlike to shortest path algorithm which focused only on some links (shortest link) which lead to some links to become overloaded, leaving others unused. This had an influence on the flow and then on the error rate.

B. Packet Loss Rate
Fig . 6 illustrates the packet loss rate as a function of number of source VPNs for each routing technique E-TSR, TSR and SPR. As shown in this figure, with low number of source VPNs average error rate with E-TSR and TSR become more frequent due to the fact that the routing techniques must search new link every time there is a new source VPN. It can be noticed that when the number of VPN sources increases to 12 sources, SPR and E-TSR have almost the average error rate, 9.6 × 10 −4 and 11.5 × 10 −4 , respectively.
Furthermore, as the number of source VPNs increase, the gap between SPR, TSR, and E-TSR increases and as we can see E-TSR provides less packet error rate. Indeed, with a shortest path routing technique all the traffic takes the same path which lead to a huge traffic on some links and then more error rate. However, with distributed traffic routing technique the traffic flow is sent over the network moderately over all links.
Furthermore, using a traffic distributed technique, packet have low chance to enter on overloaded queues, thus dropping packets will minimized and rejecting packets will be decreased. On the other site, with shortest path routing technique, there is a high probability the traffic takes the same path leading to a queue overload and then increase rejected packets.

C. Average End-to-end Delay
In this section we examine the mean delay to send a packet from one source to a destination. This delay is given according to the number of source VPNs on the network. Fig. 7 shows that the delay of three heuristics E-TSR, TSR, and SPR increases linearly with the number of source VPNs.
Moreover, Enhanced Traffic Split Routing algorithm exhibit a brief variation on delay compared to the Shortest Path Routing Algorithm. From Fig. 7 we remark that SPR has a delay around of 30 ms with 6 source VPNs. By increasing the number of source VPNs we see that E-TSR offer less end-toend delay compared to TSR and SPR. In fact, E-TSR and TSR look increase on logarithmic fashion compared to SPR.
It is clear to see that in the case of 16 source VPNs the average end-to-end delay is equal to 25.32 ms for E-TSR algorithm whereas it is equal to 27 ms for TSR algorithm, and 39 ms for SPR algorithm. Table II gives an overview of the measured values related to the mean delay for three routing algorithms. The gap on the end-to-end delay increases as number of source VPN increase. Our observations, for instances imply that when the network deploy a traffic distributed technique offer more chance of using a less queue memory which means packets sent on different links have more chance of going through small queues and therefore a small delay variation as shown on Fig. 8.

V. CONCLUSION AND FUTURE WORK
In this paper, we proposed an enhanced traffic split routing algorithm. Such algorithm is compared with shortest path algorithm. Simulations were performed using NS-2 to analyze the functionality and performance of the proposed algorithm in terms of average data rate, packet loss rate, and average end-to-end delay.
The results show that Enhanced Traffic Splitting Routing algorithm provides least values on packet loss rate and average end-to-end delay compared to Shortest Path Routing and legacy Traffic Splitting Routing algorithm.
Also, simulation results show that TSR provides better performance in term of average data rate. So, it is concluded that enhanced traffic split routing algorithm has the capability to provide better low packet loss rate and data rate by using 72% of network links compared to shortest path routing algorithm which uses only 44% of network links.