Comparative Analysis of Traffic-Reduction Techniques for Seamless CAN-Based In-Vehicle Network Systems

Abstract

Due to the benefits of better bandwidth and reliability, the automotive industry is moving towards Ethernet-based in-vehicle network (IVN) systems as the number of onboard electronic control units has increased. Considering that before long the well-known controller area network (CAN) will still be considered a standard protocol, our earlier work introduced a high-availability seamless redundancy (HSR)-based Ethernet network architecture that provides IVNs with fault tolerance, called seamless CAN. However, HSR is known for its redundant traffic generated for fault tolerance, which is a disadvantage in bandwidth-limited IVN systems. Therefore, in this paper, we propose a traffic-effective architecture for seamless CAN-based networks. We compared the proficiency of different traffic-reduction approaches as they were applied to our proposed architecture. Extensive simulation results showed that our proposed solution could reduce up to 54% of the total network traffic compared to a conventional architecture while still being able to guarantee a high level of fault tolerance.

1. Introduction

With the recent popularization of electric vehicles, their owners can now obtain access to additional on-board functionality. For example, there are more kinds of infotainment media, autopilot, and supercharging capabilities. Undoubtedly, these vehicular advancements also present technological challenges. In fact, the number of electronic components inside modern vehicles has risen dramatically over the past few years, as they involve electronic control units (ECUs) and a wide range of different sensors. As a result, there is a rising demand for in-vehicle network (IVN) capacity for the increased data traffic generated by these on-board electronics. In addition, this also leads to the complication of the cable harness and possibly heavier system weights. Therefore, it can be seen that modern cars face numerous challenges as they are being developed to provide more utilities with nearly hundreds of electronic parts [1].

However, the present IVN protocols seem to be inadequate for recent traffic-demanding network systems. In fact, the main structure of current IVNs is a combination of different protocols, for example, media-oriented systems transfer (MOST), FlexRay, a controller area network (CAN), and their corresponding inter-protocol switches. Additionally, even if CAN outperforms other protocols in terms of utilization, its speed and bandwidth limits have been reached due to the desire for higher data rates [2]. Due to these limitations, there may be a higher likelihood of in-vehicle network outages or malfunctions, which could be directly related to accidents, including those involving fatalities. In fact, as more electronic gadgets have been installed inside cars, there have been various reported accidents caused by sudden unexpected acceleration (SUA) [3]. As a solution, Ethernet-centric protocols have a superior data rate performance and may be a potential contender for future IVN systems. However, conventional Ethernet offers limited strategies for redundancy solutions in the event of data transmission failures.

In reality, several cutting-edge schemes for next-generation Ethernet-based network protocols for automobiles have been recently proposed. For example, in order to increase the reliability of Ethernet-centric networks, the Time-Sensitive Networking (TSN) Group has established the IEEE 802.1CB standard, which makes copies of a data frame and sends them using multiple routes [4]. This is probably going to be a popular protocol for future IVN systems. Similarly, this idea has also been applied by high-availability seamless redundancy (HSR), which is standardized in IEC 62439-3:2016 [5]. In fact, the HSR protocol is the foundation of the IVN architecture presented in [6], where seamless redundancy was added to the standard Ethernet-centric architecture featuring a backbone ring and four sub-rings for different function domains of commercial cars.

However, CAN is still regarded as a standard protocol because a large number of IVN systems already have it installed throughout automobile industry development efforts. For this reason, we recently proposed seamless CAN [7], a novel CAN algorithm with fault tolerance. The protocol applies the HSR redundancy method to CAN and can be viewed as a temporary solution until future IVN techniques are developed. However, in ring-of-ring networks, the standard HSR protocol produces an enormous amount of redundant traffic that is not always useful. In addition, the data traffic for entertainment media in IVNs does not have to be fault-tolerant. Consequently, network performance might degrade due to congestion and latency. Therefore, in this paper, we propose a seamless CAN-based IVN design that decreases the redundancy overhead where applicable. In detail, our main contributions are summarized as follows:

• By carefully studying the recently proposed seamless CAN protocol as a potential prospective candidate for modern IVN solutions in terms of its working principles and limitations, we proposed an architecture for effective network traffic usage where unnecessary infotainment data frames are not made fault-tolerant to make room for more critical traffic.
• Through an in-depth comparative analysis among several well-known traffic-reduction techniques for the original HSR protocol, we were able to further minimize the network traffic in our proposed architecture by choosing the suitable methods applicable to seamless CAN protocol. The compatible techniques are versatile and can be applied to any type of traffic while still being easy for implementation.
• We implemented the proposed architecture and its traffic-reduction techniques using the OMNeT++ simulation framework [8]. Next, we conducted a comparative assessment to evaluate the performance of different traffic-reduction techniques for the conventional and the proposed traffic-effective network architectures. Through in-depth discussions and analyses, we further demonstrated the feasibility of seamless CAN for various IVN system designs.

The rest of this paper is organized as follows: In Section 2, we review relevant well-known traffic-reduction techniques for the original HSR protocol. Section 3 provides readers with background knowledge in terms of the novel seamless CAN protocol and related technologies, including its relevant traffic-reduction techniques and our proposed traffic-effective architecture. In Section 4, we present the comparative analysis that was conducted through simulation amongst different traffic-pruning schemes. Finally, the concluding remarks are provided in Section 5.

2. Related Work

Because one of the main drawbacks of the traditional HSR protocol is the excessive amount of duplicated traffic generated to guarantee seamless redundancy, it is very important to study several well-known existing traffic-reduction techniques to determine the applicability of the HSR protocol for in-vehicle network systems. In this section, we first revisit the fundamental concepts of the HSR protocol, its network components, applicable topologies, and potential operational issues. We then discuss several techniques proposed in the literature to reduce unnecessary redundant traffic in HSR. These techniques can be classified into two main categories: traffic-filtering-based techniques and predefined path-based techniques.

2.1. HSR Fundamental Concepts

High-availability seamless redundancy (HSR) is standardized in IEC 62439-3:2016 [5] by the International Electrotechnical Commission. In switch Ethernet networks, it is a redundancy protocol, particularly for time-sensitive applications that require zero switchover time in the event of a network failure. It is important to note that HSR implements redundancy at Layer 2, or the data link, of the OSI model, and because it acts as an Ethernet interface for higher levels, it is compatible with protocols that use the IEEE 802 link layer. HSR employs a single network, as opposed to the Parallel Redundancy Protocol (PRP), which duplicates the whole network to create redundancy. More specifically, HSR primarily employs a ring topology in which there are two paths between each pair of nodes, as opposed to employing a different network to establish a second physical connection from the source to the destination. Two duplicate copies of each frame supplied by a node are simultaneously injected clockwise and counter-clockwise into the ring over two different Ethernet ports. As long as one path is still operational, HSR should theoretically be able to provide zero recovery time even in the event of a node or link failure. In an HSR network, there is no dedicated switch because every node in the ring has a switching function.

Figure 1 provides a representation of the HSR operation. In this illustration, the network’s nodes are all serially connected to their two adjacent neighbors, resulting in ring topology. A frame is sent by the “Source” node by inserting two copies of it (i.e., the “A” and “B” frames) across two Ethernet ports in the ring. A node will advance a frame to another port if the node is neither its source nor its only destination after receiving it unless the node has already forwarded that frame in the same way. However, even if a node receives a unicast frame (also known as an “A” frame) for the first time, its “Destination” node does not transmit the frame. Instead, the HSR tag is deleted before the frame is sent to the top layer. In addition, when the second duplicate of that frame (the “B” frame) reaches the “Destination” node, it will just be thrown away. Except for the node that injects the package into the network, every node will forward the frame under multicast or broadcast scenarios.

Figure 1. An example of unicast in a single-ring topology [4].

A node must encapsulate the information from the upper layer inside an HSR frame before sending any frames to the network, as shown in Figure 2. In contrast to the original Ethernet frame, an HSR tag is added between the payload and the link-layer header (MAC addresses and, if applicable, a VLAN tag). The HSR tag contains four fields and is used to identify duplicate data:

• “PT” is the EtherType and is used to identify a frame of HSR type;
• “Path” is reserved for experimental purposes;
• “LSDU size” is the size of LSDU;
• “Sequence number” is used to enumerate the frames.

Figure 2. An illustration of the HSR frame format.

The subsequent node types are defined by the HSR protocol:

• Doubly attached node for HSR (DANH). An HSR-capable switching end node with two HSR ports sharing the same media access control is known as a DANH node (MAC). A single interface is formed by these two ports. The source DANH replicates the frame received from the upper layers into two frame copies, adds the HSR tag to each copy, and sends them over both ports. In the event of a failure-free transmission, the destination DANH receives two copies of the identical frame, one from each port;
• Redundancy box (RedBox). Single-attached nodes (SANs) cannot be directly linked to an HSR ring because they lack the HSR tag support and HSR-forwarding functionality. To join SANs to the HSR ring, a RedBox is utilized. Like DANH nodes, RedBoxes operate as proxies for any SANs that access them and forward the frames over the ring;
• Quadruple port device (QuadBox). HSR rings are linked together using QuadBox nodes. Each pair of the four HSR ports on a QuadBox, which are separated into two groups of two and connected by an interlink, uses the same MAC. QuadBoxes function simultaneously as HSR nodes for both rings.

The principal application of HSR is for ring topologies and both single-ring and connected-ring networks to employ the HSR protocol. HSR rings may be joined using QuadBoxes to provide more intricate network topologies. A QuadBox passes frames without modification to another ring by forwarding them over each ring. Two QuadBox nodes are employed to avoid a single point of failure, even though one QuadBox is adequate to carry all traffic in a fault-free network.

2.2. HSR Traffic-Reducing Techniques

While the HSR protocol does not have any issues with unicast frames in single-ring networks, it has a major disadvantage when it comes to connected-ring networks. In fact, HSR generates too much unnecessary unicast traffic for redundancy, which is mainly caused by forwarding frames into either all DANH or QuadBox rings and by doubling and circulating frames throughout the network rings. This drawback lowers network performance and might lead to possible congestion and delay. As a result, a number of methods for decreasing HSR unicast traffic have been suggested to address the HSR issue [9]. The following two categories can be used to group these techniques: traffic-filtering-based methods and predefined-path-based methods. In HSR networks, approaches based on traffic filtering reduce unicast traffic for rings, preventing the traffic from doubling and recirculating in the rings. In contrast, two distinct paths are established for each connection pair of terminal nodes in predefined-path-based approaches. Rather than duplicating and forwarding frames to all areas of the networks, as in the traditional HSR protocol, these dual paths are then employed to send unicast packets between each connecting pair of nodes.

First, we will discuss several traffic-reduction solutions based on traffic-filtering-based techniques. By addressing any or all of the HSR-related problems, traffic-filtering-based approaches minimize duplicate traffic. There are a few popular traffic-filtering-based strategies, including the following:

• Quick removing (QR) [10]. When every node in a ring has received one copy of a frame and is about to receive a second copy, QR is used to remove circulated and duplicated frames from the ring. QR may be used for any traffic, including unicast, multicast, and broadcast, in any network architecture; therefore, it is appropriate for ring or connected ring topologies. The most straightforward method for reducing duplicate traffic in HSR networks is QR. This method is simple to use. However, QR’s biggest drawback is that it does not reduce traffic for unused rings. Network unicast frames are forwarded into all rings of HSR networks using the QR method, including QuadBox rings and DANH rings that do not include the frames’ destinations and do not carry unicast traffic from the source to the destination.
• Traffic control (TC) [11]. Similar to QR, TC is a straightforward technique to eliminate duplicated and circulated frames from a ring. In the case of terminal rings, the TC method is used. One of the ring’s nodes is chosen by TC to serve as the traffic control node. Duplicated frames will be discarded based on decisions made by the traffic control node. TC is a straightforward method, like QR, for reducing duplicate traffic in HSR networks. TC is applicable for all types of traffic, including broadcast, multicast, and unicast, in ring topologies. TC strategy does, however, have significant drawbacks. TC cannot reduce traffic for unused rings as QR can. Network traffic may be redirected into any network ring thanks to TC. In contrast with QR, which applies to both DANH and QuadBox rings, the TC technique only applies to DANH rings.
• Port locking (PL) [12]. By filtering and removing traffic for idle DANH rings, PL lowers redundant unicast traffic in HSR connected-ring networks. Without using network control messages, the PL technique allows for the network to progressively learn the position of the target node. It then locks the entry ports of the relevant DANH rings to prune those that do not include the destination node. By filtering unicast traffic for unused DANH rings, the PL method lowers redundant unicast traffic in HSR networks. The PL technique avoids increasing control overhead in HSR networks because it does not employ control messages to filter traffic. The PL technique does, however, have significant drawbacks. The PL method does not eliminate unneeded QuadBox rings from unicast traffic. Additionally, all QuadBox rings continue to replicate and transmit unicast packets.
• Enhanced port locking (EPL) [13]. The PL methodology has been improved with the EPL method. In contrast with the PL, which only prunes unicast traffic for DANH rings, the EPL technique filters unicast traffic for both QuadBox rings and DANH rings. In comparison to the PL technique, the EPL strategy enhances traffic performance in HSR networks. The EPL method prunes unicast traffic for both underutilized QuadBox rings and unused DANH rings. As a result, it reduces unicast traffic more effectively than the PL technique. Furthermore, the EPL technique, like the PL approach, does not prune unicast traffic using control messages; as a result, there is no additional control cost in the networks.
• Filtering HSR traffic (FHT) [14]. In comparison to the PL technique, the EPL strategy enhances traffic performance in HSR networks. The EPL method prunes unicast traffic for both underutilized QuadBox rings and unused DANH rings. As a result, it reduces unicast traffic more effectively than the PL technique. Furthermore, the EPL technique, like the PL approach, does not prune unicast traffic using control messages; as a result, there is no additional control cost in the networks. The FHT method prunes unicast traffic for unused QuadBox rings in addition to underutilized DANH rings. It also takes circulating and repeated traffic out of the active rings. These characteristics elevate the FHT method to the top of the list of HSR traffic-reduction strategies.

By forwarding unicast traffic across two different predetermined pathways as opposed to duplicating and flooding the traffic as in the conventional HSR protocol, predefined path-based approaches decrease duplicate traffic in HSR networks. The following are a few predefined path-based methods:

• Optimal Dual Paths (ODP) [15]. In HSR networks, ODP creates dual pathways for each connection pair of terminal nodes, such as DANH nodes and RedBox nodes. The following criteria are used to identify the dual paths: ideal connection metrics and an absence of shared nodes. Dual pathways are automatically established between all the terminal nodes in a network. Each terminal node must be aware of the measurements for all of the network links in order to identify the dual pathways with the best metrics. As a result, the ODP technique creates a table that includes metrics for every network link. The ODP technique finds all possible pathways for each connection pair based on that table, arranges them according to ascending path metrics, and identifies dual paths that meet the aforementioned requirements. By sending the redundant unicast data over two predetermined pathways rather than duplicating and flooding the entire network, ODP dramatically decreases redundant unicast traffic when compared to the normal HSR protocol. Additionally, dual pathways offer the best connection metrics, enhancing the network quality of service. The fundamental disadvantage of ODP is that it must create and maintain the link information table, adding to the network’s management cost.
• Dual Virtual Paths (DVPs) [16]. For every connecting pair of terminal nodes in a network, DVP creates two specified pathways. The DVP technique, in contrast with ODP, finds the dual pathways by sending and receiving control messages, including path selection and path confirmation messages. ODP finds the dual paths based on the network’s connection information. In the DVP setup procedure, these dual pathways will be automatically constructed. After a learning task, such a procedure merely renders all of the intermediate QuadBox nodes as “smart”. The QuadBox nodes are, therefore, able to forward a unicast frame to the intended recipient over just one of their ports, as opposed to flooding and duplicating the message through all of their ports (apart from the port on which it was received), as in the case of the traditional HSR protocol.
• Ring-Based Dual Paths (RDP) [17]. In contrast with the typical HSR protocol, which duplicates and floods data, RDP sets two predetermined paths for each connection pair and uses them for point-to-point communications. RDP, however, creates dual routes for each connection pair of terminal rings, in contrast with ODP and DVP, which do so for each connection pair of terminal nodes (DANH rings). Dual pathways used in the RDP technique are ring-based dual paths. The RDP method uses the dual pathways that have already been created between the source DANH ring and the destination DANH ring to transfer unicast frames from a source node to a destination node.

It can be seen that based on how they reduce traffic, these popular schemes can be divided into two groups: those that use traffic-filtering techniques and those that use predefined paths. Each method has benefits and drawbacks. The specific application and trade-offs determine which HSR traffic-reduction technology should be used. Traffic performance, network topology, traffic volume control, and storage needs are a few of the goals that need to be considered.

3. Background

In this section, we provide a review of our recently proposed fault-tolerant algorithm for the conventional CAN protocol: seamless CAN. Many auto-plant facilities have been set up to function with CAN because it has been used as a foundational protocol for IVN ever since it was developed. Therefore, it may take a while for current IVNs to develop into fully Ethernet-centered networks, indicating the necessity to address CAN’s shortcomings. In the following section, we first revisit the fundamental operating concept of seamless CAN, and we then discuss many possible solutions for traffic-reduction techniques that can be applied with it.

3.1. Seamless CAN

Seamless CAN is a novel fault-tolerant scheme based on HSR ring topologies for IVNs [7]. Because each traditional CAN “bit” from its source node is enclosed inside an HSR frame (i.e., it becomes the payload of that HSR frame) and its duplicates are then put into the ring, following the HSR principle, our proposed technique is also backward compatible with CAN-based systems. It should be noted that each HSR frame only contains one bit (i.e., either a “0” or a “1”) rather than a full CAN frame (see Figure 3) because the CAN bus is being replaced by an HSR ring, and the CAN controllers inside CAN nodes are sending the bits serially to the bus. We present the idea of a seamless CAN node with three separate interfaces in order to connect with the CAN node and its two nearby seamless CAN nodes, as given in Figure 4.

Figure 3. Comparison between a (a) standard HSR and (b) seamless CAN frame.

Figure 4. Three interfaces of a seamless CAN node [7].

In terms of an operational concept, a seamless CAN node will embed information into the payload section of an HSR frame and send it out in two ways via two distinct interfaces on the HSR side when a bit is received from the CAN side’s interface. Additionally, the sequence number of a new frame will be recorded in the node memory when it arrives at either interface on the HSR side, and the bit information will be extracted and sent to the CAN node via the interface on the CAN side. Similar to HSR, if a frame from the HSR side has a sequence number that already appears in the node’s record, it will be deleted.

It is clear that the frame transmission activity is comparable to HSR’s function once the CAN signal has been enclosed inside the seamless CAN frame and delivered in two copies when there is no failure in the network. However, the network might have a connection failure. Nodes on the CAN side are invisible to actions occurring on the seamless CAN side, therefore, even when only one copy of the frame is sent to the target node, the CAN node continues to function normally. While it is clear that both CAN and seamless CAN frames can become corrupted, seamless CAN frames are less likely to have faults because they can be recovered via the other path. When a faulty frame is discovered, nodes in traditional CAN networks continuously attempt to resend the frame. Additionally, error correction for encapsulated CAN frames is supported by seamless CAN.

3.2. Traffic-Reduction Techniques for Seamless CAN

Although the aforementioned seamless CAN scheme can be used in conventional CAN-based systems and can provide seamless redundancy for data-critical IVNs, it is not fully optimized because HSR by its very nature generates too much redundant traffic to achieve fault tolerance. Therefore, we observe that several well-known traffic-reduction techniques can be applied to seamless CAN-based IVNs to more effectively utilize the network throughput while still being able to provide fault tolerance to time- and data-critical network packets.

First, it should be noted that while there are many traffic-reduction schemes for the HSR protocol, not all of them can be used in a seamless CAN-based network. This is because seamless CAN is a fault-tolerant version of a CAN network, which is mainly functional by using a shared bus (i.e., CAN bus) among CAN nodes. In fact, traditional CAN nodes communicate by listening and transmitting frames to the shared bus, which can be considered as broadcasting messages. Therefore, only traffic-reduction techniques that are suited for decreasing broadcast traffic in the original HSR network can be used in a seamless CAN-based network. According to [9], it can be seen that among various traffic-reduction techniques, only two of them are applicable to networks with a seamless CAN configuration:

• Quick removing (QR);
• Traffic control (TC).

A comparative analysis of QR and TC among other techniques can be found in Table 1. It can be seen that QR and TC are the two simplest approaches to decrease redundant traffic in HSR networks as well as seamless CAN networks. Indeed, QR and TC are very versatile as they are able to reduce any type of traffic in the network, including unicast, multicast, and broadcast, for any network topology. In addition, they are both straightforward to implement and do not require any control messages in order to operate, thus avoiding possible protocol overhead.

Table 1. Comparison between different traffic-reduction techniques for seamless CAN-based networks, where “Y” denotes “Yes”, “N” denotes “No”, and “-” is “not applicable.”

Characteristics	Traffic-Filtering-Based					Predefined Path-Based
	QR	TC	PL	EPL	FHT	ODP	DVP	RDP
Traffic filtering for DANH rings	N	N	Y	Y	Y	-	-	-
Traffic filtering for QuadBox rings	N	N	N	Y	Y	-	-	-
Removing duplicate traffic	Y	Y	N	N	Y	Y	Y	Y
Learning MAC addresses	N	N	Y	Y	Y	-	-	-
Using control messages	N	N	N	N	Y	N	Y	Y
Configuration to setup dual paths	-	-	-	-	-	Node	Node	Ring
Using network topology information	-	-	-	-	-	Y	N	N
Type of filtered traffic	Any	Any	Unicast	Unicast	Unicast	Unicast	Unicast	Unicast
Applicability to Seamless CAN	Y	Y	N	N	N	N	N	N

3.3. The Proposed Traffic-Effective Architecture

First, we describe the typical network architecture in a typical commercial automobile, which normally consists of four main function domains: power, chassis, body, and infotainment [18]. Each domain consists of seamless CAN nodes that are attached to regular CAN nodes, and the four domains are connected using HSR QuadBoxes on the HSR side, collectively forming a main backbone ring. The seamless CAN nodes can be used to create a connected-ring kind of network because they are entirely compatible with the traditional HSR protocol. Seamless CAN nodes contain two Ethernet interfaces, much like the original HSR DANH devices, to maintain two transmission channels to every other node in the network, even if they are not in the same sub-ring. Because of this, the seamless CAN-based network architecture can be implemented without the use of any new hardware on the present IVN system. An illustration of a seamless CAN-based network can be found in Figure 5.

Figure 5. Example of a seamless CAN-based network architecture with four domain functions in a car.

However, in spite of the fact that the aforementioned seamless CAN technique can be used in conventional CAN-based systems and can offer seamless redundancy to data-critical IVNs, it is not completely optimal since HSR by its very nature creates too much redundant traffic to accomplish fault tolerance. This problem is mainly induced by the following reasons:

• Frames are sent into all DANH rings in addition to the one containing its destination;
• Frames are sent to all QuadBox rings;
• Frames are duplicated and circulated across all rings.

Given the excessive redundant traffic generation of seamless CAN, we propose a traffic-effective network architecture that is compatible with current IVNs and is able to successfully reduce the number of traveling network frames. The proposed traffic-effective network structure is depicted in Figure 6.

Figure 6. Example of an enhanced seamless CAN-based in-vehicle network.

Compared to the traditional four-domain architecture, the traffic-efficient one has the following differences (although both may look similar in the way that they share the same four-QuadBox design):

• First, we noticed that in addition to parts of an IVN system that are in charge of transmitting time-critical information, such as sensor data and a brake controlling signal, there are also network components that communicate non-critical information (e.g., radio and media playing services). Obviously, they do not play an important role in ensuring the safety of car users because their exchanging information is not crucial. In addition, it is noted that these infotainment parts of the network consume a large amount of data traffic as they are bandwidth-demanding for high-quality audio and video transmission;
• Therefore, in the proposed traffic-efficient network architecture, we decided not to implement fault tolerance for non-crucial parts of the network. Specifically, because entertainment units can be collectively modeled as any one of the four QuadBox rings (as depicted in Figure 5), we can replace the seamless CAN nodes with SAN nodes. In addition, the SAN nodes are connected using an HSR RedBox, which is a device that enables non-HSR devices to join the ring-type HSR network. In this manner, SAN nodes (e.g., media players and camcorders) are still able to communicate with the rest of the network while there is no need to provide seamless redundancy for them. Ultimately, the other parts of the entire network remain the same.

In summary, the proposed traffic-effective architecture has the following advantages when it comes to designing, utilizing, and providing fault tolerance to the data traffic of the network:

• First, by making use of the novel fault-tolerant seamless CAN method, we ensure that mission-critical network frames always have two redundant paths between the source and destination nodes in the error-free scenario. In the case of faulty network links, there would still be no delay in the delivery process of the data frames as long as one path is still available;
• Second, the proposed architecture removes the fault tolerance capability for the infotainment components to reduce unnecessary traffic generated by them as they are the parts that may greatly consume bandwidth in the network;
• Finally, by using either the QR or TC traffic-reduction technique, we are able to further decrease a substantial amount of redundant traffic in the other parts of the network without compromising their seamless redundancy function.

4. Comparative Evaluation of Traffic-Reduction Techniques

In this section, we provide simulation results that compare the effectiveness of the proposed network architecture under many scenarios of the no traffic-reduction technique applied, QR technique used, and TC technique used. All simulation works were conducted using the OMNeT++ discrete simulation framework [8]. It should be noted that for each architecture, we compared the number of total network traffic (in frames) against the number of sent frames as well as the size of each sub-ring (in nodes) in the network. Without loss of generality and for brevity, we considered that four sub-rings had the same number of intermediate nodes. A summary of our simulation setup can be found in Table 2.

Table 2. Simulation parameters for two scenarios, where Scenario 1 is the performance result against the numbers of sent frames, and Scenario 2 is the performance result against sub-ring sizes.

	Parameters	Value
Scenario 1	Number of sent frames	$\{100,1000,2000,5000\}$
	Sub-ring size	20
Scenario 2	Number of sent frames	2000
	Sub-ring size	$\{5,10,20,30\}$

4.1. For Conventional Architecture

First, we show the performance of the unmodified conventional architecture (as given in Figure 5) in terms of the number of total generated frames. More specifically, in this scenario, we set the number of seamless CAN nodes in each of the four sub-rings to 20. During the simulation period, a source node tried to generate and inject different numbers of broadcast frames into the network. At the end of each simulation run, the total generated traffic in frames across the entire network was reported. The result for this scenario is depicted in Figure 7.

Figure 7. Traffic-reduction performance against numbers of sent frames for the conventional architecture.

Generally speaking, to maintain the seamless redundancy of the network, there were many more frames traveling in the network compared to the actual number of generated frames. For example, without applying any traffic-reduction schemes, while there were only 5000 frames broadcasted by the source node, nearly 900,000 frames were created by the protocol for fault tolerance purposes. In contrast, the TC approach slightly reduced the broadcast traffic by 31%, at roughly 600,000 generated frames for 5000 original frames. Ultimately, the QR scheme reduced the traffic by 41%, and this was the best result out of the three experimental schemes.

Next, we compared the performance of the same conventional architecture among different numbers of seamless CAN sub-ring nodes. In this setting, each sub-ring in the network was supposed to equally have 5, 10, 20, and 30 nodes, and the number of broadcast frames was set to be 2000 for every simulation run. The total network traffic measured in frames was recorded at the end of each simulation. The final result for this scenario is given in Figure 8. It can be seen that, similar to the first scenario, the network generated much more traffic as it had more immediate nodes in each ring. For the largest size of the network (i.e., 30 nodes in each sub-ring), more than 500,000 generated frames from 2000 original data frames were required for the system to be considered fault-tolerant for the conventional architecture without any traffic-reduction approaches. In contrast, the TC approach only produced about 350,000 frames, and the QR approach reduced about another 50,000 frames, to as low as roughly 300,000 frames.

Figure 8. Traffic-reduction performance against different sub-ring sizes for the conventional architecture.

4.2. For the Proposed Architecture

To evaluate the performance of our proposed architecture (as given in Figure 6), we implemented it in the same simulation environment as the conventional architecture. We also demonstrated the performance of different traffic-reduction techniques for the proposed architecture in terms of total network traffic in frames. It should be noted that, in the proposed architecture, because there were only three sub-rings compared to four in the conventional one, we attached 20 SANs to the RedBox to maximize fairness between the two schemes. Apart from this difference, all the simulation settings, including the number of sent frames and the broadcasting mode, remained the same as in the previous scenario.

The simulation results are given in Figure 9. It can be seen that the proposed architecture used nearly 200,000 frames fewer than the conventional one of roughly 900,000 frames. In this case, the TC approach decreased the traffic to around 53% of the original number of frames without any reduction schemes. QR was still the most efficient method as it was able to further decrease the amount of traffic to almost 45.5%.

Figure 9. Traffic-reduction performance against numbers of sent frames in the proposed architecture.

Next, we show the performance of our proposed architecture in terms of different sub-ring sizes. We kept the same settings as with the conventional architecture except for the fact that the RedBox device had the same number of SANs attached to it as the number of seamless CAN nodes in each scenario (i.e., 5, 10, 20, and 30 SANs). The final simulation result is given in Figure 10. To begin, it can be observed that the performance of both the TC and QR approaches was relatively similar to the previous experiment in terms of network traffic. In other words, without a suitable reduction technique, the number of network frames was quite large because they had to circulate over the network to guarantee seamless redundancy with zero delays. The TC approach reduced nearly 30% of the traffic, and the QR approach was able to further diminish another 10%, down to as low as 50% of the original amount of traffic.

Figure 10. Traffic-reduction performance against different sub-ring sizes in the proposed architecture.

4.3. Evaluation

From the given simulation results, we have shown how the TC and QR traffic-reduction schemes can benefit seamless CAN networks, regardless of the architecture to which they were applied. To better demonstrate the efficiency of the proposed architecture combined with traffic-reduction techniques against the conventional one, Table 3 recapitulates the relevant results in numbers. Note that because the performances of the TC and QR approaches were nearly the same for both scenarios with different numbers of sent frames and sub-ring sizes, we only present the former results for brevity.

Table 3. Traffic-reduction performance against numbers of sent frames for the conventional and proposed architecture, where “-” denotes no reduction technique being used.

No. of Sent Frames	Conventional			Proposed
	-	TC	QR	-	TC	QR
100	17,600	12,150	10,382	13,562	9366	8000
1000	176,000	121,550	103,822	135,616	93,656	80,000
2000	352,000	243,090	207,645	271,232	187,312	160,000
5000	880,000	607,730	519,112	678,080	468,280	400,000

For better clarity, Table 4 presents the reduction percentage of different schemes against the conventional architecture without using any traffic-reduction technique. It is clear that while the most effective technique was the QR approach as it was able to reduce up to around 41% of the total generated traffic for the conventional architecture, our proposed architecture diminished a certain number of network frames with a reduction effectiveness of up to 54%. Therefore, it is safe to conclude that we can minimize the network traffic the most by utilizing both the QR approach and our proposed IVN architecture. This is undoubtedly important in data-critical networks as the network traffic is greatly reduced while the level of fault tolerance capability is still guaranteed.

Table 4. Traffic-reduction performance against the numbers of sent frames for the conventional and proposed architecture.

No. of Sent Frames	Conventional		Proposed
	TC	QR	TC	QR
100	30.97%	41.01%	46.78%	54.55%
1000	30.94%	41.01%	46.79%	54.55%
2000	30.95%	41.00%	46.79%	54.54%
5000	30.94%	41.01%	46.78%	54.55%

5. Conclusions

In this paper, we proposed a traffic-effective seamless CAN-based network architecture for IVN solutions. By only providing fault tolerance to essential parts of the network in combination with relevant HSR-based traffic-reduction techniques, our proposed solution was demonstrated to significantly reduce the total amount of generated network traffic while still being able to maintain the same level of seamless redundancy. The simulation results also showed that quick removing (QR) was the best reduction technique for seamless CAN-based networks. Therefore, our proposed architecture is a suitable solution for CAN-based network systems where time- and data-critical packet transmissions are required.

Later, our future work may include real-world implementation to demonstrate the applicability of seamless CAN solutions to commercial automobiles. In addition, as we are currently only making use of well-known HSR-based traffic-reduction techniques, we may devise more effective techniques specifically designed for a seamless CAN network to better utilize the network traffic by further reducing unnecessary data frames.