Efficient Bloom Filter-Based Routing Protocol for Scalable Mobile Networks ^†

Abstract

Non-geographic routing protocols are inefficient when applied to large-scale mobile networks composed of hundreds of nodes. On the other hand, geographic routing protocols have the disadvantage of needing a location sensor. The goal is to address the challenges of efficient content retrieval and routing scalability in NDN-based networks by leveraging the benefits of both NDN and Bloom Filter technologies. In this article we propose a routing protocol for mobile networks, which is scalable to networks composed of hundreds of nodes. The protocol does not require any localization equipment and is adapted for devices with limited memory and/or processing resources. This goal is achieved through the use of Bloom Filters to efficiently store and spread topological information. In the methodology followed, nodes do not forward messages with topological information to other nodes. To make the process efficient, each node aggregates the topological information it receives from its direct neighbors with its own and only the result of this operation is transmitted to the remaining nodes. Several simulations were carried out in the Qualnet network simulator in order to validate the algorithm proposed by the Hybrid Routing Algorithm with NDNs (HRAN). The obtained results were compared with other non-geographic protocols for mobile networks. HRAN seems to be a routing protocol designed for MANETs, utilizing Bloom Filters to manage topological information. A Bloom Filter is a data structure used to test whether an element is a member of a set. It uses a bit array and multiple hash functions to determine if an element is present in the set. This type of data structure allows storing a large amount of binary information in an efficient way, reducing the resources required by the routing protocol.

1. Introduction

Ad hoc mobile networks, also known as mobile ad hoc networks, have become more important due to the widespread use of wireless technology in recent years (MANETs). Because of its unique properties, such as its dependence on a scarce energy supply and/or its susceptibility to rapid topological changes, this class of networks has been the focus of several investigations. Proactive protocols, reactive protocols, and hybrids of these two types are the conventional buckets into which researchers have placed most forwarding protocols suggested for use in MANETs. Maintaining routes to all feasible destinations, regardless of network activity, is a hallmark of proactive routing techniques. In contrast, reactive protocols set up new paths only when needed by a source node. Three examples of proactive protocols are the Destination Sequence Distance Vector (DSDV), the Temporally Ordered Routing Algorithm (TORA), and the Optimized Link State Routing protocol (OLSR) (TORA). Dynamic source routing (DSR) and Ad hoc On-demand distance vector (AODV) protocols are two examples of reactive routing systems. Hybrid regimens, which combine reactive and proactive features, are another option. The Zone Routing Protocol (ZRP) is one such protocol [1,2,3,4,5].

Several studies have been undertaken comparing the different possibilities. After modelling both proactive and reactive protocols in medium-sized networks, the authors of found that the latter produced better outcomes in terms of latency and packet delivery %. These protocols, however, also provide the most unnecessary work for the network to carry out if more routes are requested. This research also confirms that proactive protocols lead to a greater number of lost packets and worse quality of first routes. In contrast, if we look at the total number of control messages, we find that this group of protocols is more effective [6,7]. In order to effectively store and transmit topological information, HRAN employs Bloom Filters (BFs). Nodes in an HRAN network do not relay information amongst themselves. If nodes receive messages containing topological information, they may conserve bandwidth by combining it with what they already know about the network via a binary OR operation. This operation’s output is the only thing sent to the other nodes. The HRAN is able to efficiently search for routes thanks to this function, which involves the dissemination of topological information. When a route has to be fixed or improved in terms of the number of nodes (NoNs), this approach may be utilised to create topologically informed zones around the route [8].

Currently, there are some solutions already proposing the use of BFs in forwarding protocols for ad hoc networks [9].

2. Methodology

Here, we will explain how the HRAN routing protocol really works. We will start with a quick overview of the method, then dive into the BFs’ temporal context. Finally, we will provide a comprehensive breakdown of the algorithm [9,10,11].

2.1. Concept

HRAN is an effort to translate the “heat trails” concept from the real world onto a representation of a network’s topology. By periodically emitting heat signals to its neighbors, each node generates a hotspot around itself. Nodes inside this zone “feel” the heat and have a hint on how to reach it. Those nodes nearer the goal “feel” the node closest to the destination, while those farther away “feel” the coldest heat channel. If a communicating node is located outside of the heat zone of the destination, it will have to randomly probe the network until it finds a node inside the heat region. At that point, consumer demand shifts from cooler to warmer climates. Once a path is determined, a heat tunnel may be built. This technique is helpful since it aids in subsequent searches for the same destination, fixes any route failures, and optimizes the NoNs that make up the route [12,13,14].

Figure 1 depicts the varying temperature gradients at HRAN’s node D, which are essential to the network’s functionality (destination). D-nodes are represented by darker nodes and have a warmer tone than lighter nodes. An image of a heat tunnel is shown there. Directed search and random search are conceptual models of the process of information exchange between nodes B and D.

A node’s “warmth” is fully represented by the adapted bandpass filter, which includes all relevant data. The BF is a small data structure that checks whether an item is in a set, however it’s possible for the check to provide false positives. A limited collection of hash functions (HFs) of sizes m and k are used to instantiate BF instances. These HFs each designate a specific location in the array m for use by a single set member. You need to execute each HF on its ID to obtain k positions and add a new member to the set. The number 1 is used for every pixel in the bitmap. To verify an item’s membership in the set, just re-enter its ID into each of the k HFs to determine where it is stored in the array. If any of these bits is set to 0, then the element does not belong to the set [15,16,17].

In the HRAN protocol, versions of these tiny binary structures are organised into gradients, where the location of the gradient relates to the distance of the No. of intermediate nodes whose nodes are positioned relative to each other. In order to disseminate topology information across the network, neighbours regularly swap these bespoke BFs with one another.

2.2. BF with Temporal Information

By enhancing the capabilities of the standard BF structure, the HRAN routing protocol will be able to determine whether an item has been recently added. A novel BF variant, the TAB (Time Aware Bloom), was developed for this purpose.

As can be seen in Figure 2, this is a pair of BFs (sticky and plain filter (P&S)) with the same size M and the same set of HFs K. To check whether the final slot has been occupied by a bit, the sticky filter is applied(δn). The plain filter contains the bits entered during the previous and previous time periods (δn || δn−1). Vectors always employ TAB filters; here, we detail how to perform insert, search, and update actions while using TAB filters.

The procedure uses the standard insertion operation on the P&S filters already present in the vector to add an element X at position I of the TAB filters. To determine the k locations, we must first execute the K HFs on the inserted element. In these roles, both BFs will have the value of 1.

When looking for the value Y in a TAB array, only the array’s simple filters are examined. The search element is hashed with the KHFs, and then the bits of each acquired location are checked in a series of simple filters to see whether they all contain the value 1. In such case, the search was successful. If this match is never discovered after going through all of the filters in the array, then the item is not found.

2.3. Functional Description

The HRAN routing protocol comprises three distinct stages of execution: building the heat layer, searching for the hottest paths, and keeping the routes in place. Each of these stages will be discussed in further depth below.

• Construction of the heat layer: During the heat layer’s initial setup, each node proactively shares its network topology data with its neighbours. Each node maintains an N-by-1 vector of TAB filters containing all of its knowledge about the network’s topology. There is a corresponding level of the heat gradient at each of these coordinates [18,19,20].

The thermal layer for node D with a 6-bit filter size is seen in Figure 3. The shown HRAN configuration used a total of three gradients (N = 3). Nodes in dark grey layers suggest that their D-nodes are in warmer scale locations than nodes in light grey regions. There is no information about the location of the D node in the TAB filter vector for any nodes that do not include it.

Each node contributes to the formation of the thermal layer by regularly broadcasting HELLO messages. Each HELLO signal has a vector of length N = 1 BFs, where the strongest colour gradient (N) is suppressed. In the HELLO message, you will find a BF that is an exact replica of the standard sender node filter, only with a temperature gradient of one to N = 1. In order to minimise data transfer during topological information propagation, regular filters are sent.

The temperature gradient of a node is ORed with the BF sent in the HELLO message using a binary OR operation. Figure 4 illustrates that the beginning of the filter copy in the message occurs at position 2, whereas the beginning of the position in the node structure occurs at position 1. This method operates on the premise that elemental information weakens with increasing distance.

• Route maintenance: Route maintenance is divided into three tasks: (1) Establish a hot tunnel. (2) Route is selected to transmit packets and restore direction as shown in Figure 5a. (3) The path optimization process iteratively optimizes the path to reduce the impact of navigation and randomness in the path discovery process as shown in Figure 5b.

Heat tunnels are created by entering the identification into the TAB filter located at position 1 in the heat gradient of the intermediate node, a heat tunnel is produced when the packet is sent. As the TAB filters are normally updated, the “heat” information for unused routes is progressively lost, but this procedure maintains it current for the routes that are really being utilised. Heat travels via a tunnel from node A to node D, as shown in Figure 6a. Again, nodes in darker grey zones have lower (warmer) gradient information about node D than nodes in lighter grey zones.

The construction of heat tunnels in frequently used corridors has a dual purpose. As can be seen in Figure 6b, the primary goal is to facilitate future route searches by other nodes. The second goal is to facilitate route repairs on a local level with little communication. Figure 6c depicts the construction of a new heat tunnel after the identification of the other path.

Route repair: When a route fails as shown in Figure 7a, repair signals (REPAIR) are repeatedly sent to different regions with increasing temperature. When the final destination is reached or no more hotspots are located, the procedure concludes. To perform repair process, a REPAIR message is sent to all nodes in broadcast node. The destination node on the hottest gradient (gradient 1) is the route that failed, the first repair area starts off as gradient 2. The repair area is progressively incremented until it reaches the N gradient. At this stage, the route is considered irrepairable and a new route discovery process is started which is demonstrated as in Figure 7b.

3. Results and Discussion

Qualnet network simulator setup has the following steps:

• Design a mobile network topology that reflects the real-world characteristics you want to simulate.
• Determine the number of mobile nodes, their initial positions, mobility models, communication ranges, and movement patterns.
• Integrate content sources and destinations based on your simulation objectives

Simulations were performed using Qualnet version 5.0.1. This simulator was chosen because it supports large-scale networks with hundreds of nodes and has a library with several protocols for MANETs already implemented, thus enabling an easy and reliable comparison. For all tests, the test conditions were as follows:

• Map size—1200 m by 1200 m;
• Maximum communication distance—150 m;
• Propagation model—two-ray;
• Simulated time—110 s;
• Knot speed—15 m/s;
• Node direction—random.

The parameters configured for the HRAN were as follows: an interval between HELLO messages of 7 s, use of 4 levels in the heat gradients, BFs size of 256 bits, a TAB filters update time of 21 s and a path improvement interval of 20 s.

3.1. Increased No. of Route Requests

The first performance test consists of measuring the No. of packets sent by the forwarding protocol to discover new routes in a network composed of 200 nodes. In this scenario, the chosen independent variable was the No. of source nodes. The target node remains the same for all tests. As can be seen in Figure 8a, the ZRP and OLSR protocols (INRIA implementation) maintain a value close to constant regardless of the No. of routes requested.

The DSR protocol requires fewer messages to be delivered than the HRAN for low values of the source node, but the HRAN’s layering and heat tunnelling technique enhances the efficiency of the routing process as the number of route requests grows. You can see the ratio of control packets to data packets transmitted in Figure 8b, with the HRAN routing protocol again providing the best results for larger route request numbers.

3.2. Increase in the NoN

The primary role of the HRAN routing protocol is to handle networks with hundreds of nodes, hence this is the focus of the first performance test. The second evaluation looks at the forwarding protocol’s message throughput as the network NoN grows. Figure 9a shows that the DSR and the HRAN protocol use less control packets than the other protocols for networks with 150 nodes. For the smaller network of 150 nodes, the DSR has a little lower value than the HRAN and the ZRP protocol, respectively.

As shown in Figure 9b, proactive approaches (OLSR and fisheye) never experience delays when requesting new routes since they always have existing routes in place. For large networks, the HRAN has lower values compared to the other routing protocols since it makes use of heat structures to direct searches for new routes.

3.3. Size of Routes

The last verification checks the route sizes utilised by the HRAN and DSR protocols throughout the simulation. Figure 10 shows the findings, which show that the HRAN protocol’s lengthier routes may be progressively reduced to values close to those of the DSR thanks to the route improvement mechanism being engaged every 20 s. These findings verify the efficacy of this approach in addressing a concern raised by HRAN users about the protocol’s partly randomised route search.

Improved network scalability, decreased resource consumption, efficient content retrieval, and adaptive routing are just some of the concrete advantages that result from using the suggested Hybrid Routing Algorithm in real-world mobile networks. There are benefits and difficulties associated with all different kinds of routing protocols. However, non-geographic protocols may have trouble scaling and may not be as efficient in retrieving material. Geographic protocols exploit location information but might be sensitive to movement. Content-centric protocols have scalability issues but favour efficient content retrieval.

4. Conclusions

Improved network scalability, decreased resource consumption, efficient content retrieval, and adaptive routing are just some of the concrete advantages that result from using the suggested Hybrid Routing Algorithm in real-world mobile networks. There are benefits and difficulties associated with all different kinds of routing protocols. However, non-geographic protocols may have trouble scaling and may not be as efficient in retrieving material. Geographic protocols exploit location information but might be sensitive to movement. Content-centric protocols have scalability issues but favour efficient content retrieval. As a solution to these issues, the suggested Hybrid Routing Algorithm merges the content-centricity of NDN with the memory efficiency of Bloom Filters in order to strike a good balance between scalability, content retrieval speed, and resource utilisation, especially in real-world mobile networks. To maximise both content-centric gains from NDN and memory efficiency from Bloom Filters, the Hybrid Routing Algorithm employs an aggregation mechanism. When it comes to storing and disseminating network topology data, the Bloom Filters are indispensable.