Evaluating Segment Routing Technology for MPLS-based Network

. Multi-protocol label switching (MPLS) has proved its strength for large network traffic since it emerged. However, due to increasing demands on network services and for a massive traffic that continually growing, Segment Routing was developed and introduced to overcome the limitation in load balancing and scalability of current MPLS. This paper examines the effectiveness of the SR technique for MPLS-based networks with Traffic Engineering (TE) via experimental work where an implementation of basic SR scenarios is emulated and tested for some comparison criteria. These criteria were based on packet analysis for both simple ICMP and flow in stable conditions and for network convergence in case of failure. Simulation results confirm the effectiveness of SR in MPLS-TE-based networks in terms of convergence and reducing overhead traffic control.


Introduction
The implementation of Multiprotocol Label Switching (MPLS) marked a significant milestone in the world of networking, introducing crucial features such as tunnelling and traffic steering.These innovations were instrumental in MPLS's success, particularly in offering secure and efficient Virtual Private Networks (VPNs) [1].However, despite MPLS's many advantages, it faced challenges related to scalability and load balancing when Resource Reservation Protocol-Traffic Engineering (RSVP-TE) was employed.These limitations prompted the need for a more robust solution.In response to the shortcomings of RSVP-TE, the networking community developed the Segment Routing (SR) protocol.SR emerged as a transformative approach, granting applications the power to control network behaviour while delivering enhanced scalability, simplicity, and ease of implementation.Spearheaded by Cisco engineers, SR evolved and gained widespread recognition, eventually becoming the official standard for SR in 2018 [2].
MPLS technology was developed in the late 1990s to improve data packet forwarding in large-scale networks [3].It introduced the concept of label switching, where routers assign labels to packets to indicate their forwarding path.MPLS networks used protocols like LDP and RSVP-TE to distribute labels and establish label-switched paths, allowing for faster and more deterministic packet forwarding.MPLS-TE enabled network administrators to control and optimize traffic flows by creating explicit paths, improving resource utilization and QoS.MPLS also introduced the capability to create secure and efficient VPNs.However, as network requirements evolved, SR was developed to offer a more versatile and efficient approach to traffic engineering and routing, addressing some limitations of MPLS.The study aims to assess if SR can address the limitations of MPLS in terms of load balancing and scalability.It compares the performance of MPLS-TE and SR-TE on various criteria, including packet analysis, convergence time, and overhead.Additionally, the research investigates the impact of SR on network convergence, analyzing if SR-TE offers faster convergence compared to MPLS-TE.It also examines the packet overhead introduced by both MPLS-TE and SR-TE.Finally, the study aims to determine the practical implications of adopting SR-TE for Internet Service Providers (ISPs) and how it could enhance their network performance and reliability.Overall, the research aims to evaluate the effectiveness, performance, and practical implications of SR in MPLS-based networks, providing insights into its viability as a solution for modern networking challenges.

Segment Routing
This section discusses the concept, architecture, operation and benefits of SR in the following subsections.

The Concept
Segment Routing is a routing technology that utilizes source routing, where packets contain encoded information about the chosen path.This path is represented as a stack, with segments serving as instructions for nodes to follow.These segments can have either a local meaning within a specific device or a global meaning across the entire network topology.For instance, a segment with local meaning may indicate the outgoing interface for a flow while a segment with global meaning can specify the complete path through the domain.To distinguish each segment, a unique value called the Segment Identifier (SID) is assigned [3].Data and control planes are present in almost every networking technology or protocol.SR supports three control plane types, which are specified in segments: • Distributed -In the distributed approach, segments are communicated through a routing protocol such as IGP Link State (OSPF, IS-IS), or BGP.
• Centralized -In this scenario, all signaling is performed by a specialized device called a controller.The controller gathers topology information via the BGP-Link State (BGP-LS) protocol, selects a path using the Path Computation Element Communication Protocol (PCEP), and disseminates it to application servers.
• Hybrid -The hybrid data plane offers a combination of distributed and centralized data planes.It is advantageous when the source and destination nodes reside in different IGP domains.

Architecture
SR is basically a network architecture that offers a versatile and scalable approach for directing traffic flows within a network by utilizing pre-defined paths.It operates on the principle of source routing, enabling the packet sender to designate the exact path that the packet should traverse across the network.The architecture of Segment Routing is composed of two principal components: the Control Plane and the Data Plane • Control Plane: Serving as the core component of the SR architecture, the Control Plane takes charge of computation and the selection of optimal paths for packet traversal across the network.Implementing a centralized controller, the Control Plane coordinates with network devices like routers and switches to instruct them on the suitable forwarding instructions for each packet.By utilizing network topology and traffic engineering algorithms, the controller determines the most advantageous path for each packet, considering factors including Quality of Service (QoS) requirements and network conditions.This path is denoted as a sequence of segments, commonly represented as IPv6 addresses.
• Data Plane: Serving as the forwarding engine within the SR architecture, the Data Plane plays a crucial role in moving packets through the network in accordance with the paths determined by the Control Plane.This functionality is achieved through the utilization of network devices, including routers and switches, which are programmed with the necessary instructions for packet forwarding.Upon transmission of a packet into the network, it is assigned a series of segments defining its intended path.The sender of the packet specifies this sequence of segments, and every network device along the route utilizes this information to determine the subsequent segment and effectively forward the packet.SR consists of two types of segments [4]: • Node Segments: These segments correspond to individual network nodes and facilitate traffic routing across the network.Each node within the network is assigned a distinct Node Segment ID (SID), which assists in identifying the node and establishing the appropriate path for packet traversal.• Adjacency Segments: These segments represent the links connecting network nodes and enable traffic routing along specific paths.Each link within the network is assigned a unique Adjacency Segment ID, aiding in link identification, and determining the optimal path for packet transmission.

Segment Routing Operations
The source node controls the flow of incoming traffic by attaching an ordered list of SIDs (Segment Identifiers) to the packet header.Initially, the first segment in the list is executed, and then subsequent segments are executed as the packet reaches intermediate destinations.This process continues until the last segment is executed.At this point, the flow either reaches its intended destination or leaves an SR domain and is routed based on the destination IP address using traditional routing methods.SR-capable nodes perform three actions on segments, which are similar to the operations performed on MPLS (Multiprotocol Label Switching) labels in MPLS networks.These Segment Routing operations are listed as follows

Experimental Works
This section reviews all aspects of the practical part of conducting the experiment, including system setup and configurations, experiment scenarios and test cases.

Experimental Setup
The hardware and software requirements for the study are as follows:

Experimental Scenario
Two main scenarios are considered in this work; MPLS-TE only scenario (MPLS-TE hereafter) and MPLS-TE with SR scenario (SR-TE hereafter).Different configurations of the traffic engineering tunnels are conducted to simulate these techniques.For testing and evaluating these scenarios, the following parameters are analyzed: • Link load, including the number of control plane messages transmitted on a link and the size of data and control packets on a link .• The tunnel's convergence and re-route time if the primary connection fails .Figure 2 illustrates the main topology of the experiment's scenarios.This includes: • 2 provider edge routers (PE1, PE2).
• Same IP address scheme is shown in the figure 2. The difference between the routers above is only in the configurations, for example: in MPLS-TE and SR-TE: PE1 has a configuration not like PE2, while P1, P2, P3, P4 and PE2 has the same configuration the difference is just in the addresses scheme.

Fig 2 Network Tolopgy
It is also worth mentioning that, in the two scenarios, all routers are supporting both MPLS-TE and SR-TE technologies.However, PE1 in the two scenarios has an MPLS-TE tunnel and SR-TE tunnel configured, while routers P1, P2, P3, P4, and PE2 have the same configuration.PE2 has an OSPF route redistribution command and a static route to the customer edge 2.

Result Analysis
This section discusses the experiment's results for the designated scenarios.Test results were measured by means of packet analysis using ICMP echo requests and the packet E3S Web of Conferences 469, 00075 (2023) ICEGC'2023 https://doi.org/10.1051/e3sconf/202346900075flow in a stable TE tunnel.In addition, the convergence of the network in case of a failure.

Packet Analysis using Simple ICMP Echo Request:
In MPLS-TE scenario, the packet transmitted from CE1 to CE2 or from PE1 to PE2 through the Traffic Engineering tunnel has a size of 118 bytes.The MPLS Header is 4 bytes long and contains the following values: MPLS Label = 17022, TTL = 255, Bottom of stack flag = 1 (indicating the last header).The overhead of a single packet is only 4 bytes because only one label is swapped on all nodes except the one connected to PE2.Due to Penultimate Hop Popping (PHP), packets on the link connected to PE2 do not have an MPLS header .In SR-TE scenario, the packet analyzed on the link between PE1 and PE2 carries the entire label stack, which is defined on the originating device.It contains three labels in the stack: 17003, 17005, and 17004.The total size of the packet in SR-TE is 8 bytes larger than in MPLS-TE due to the increased overhead of carrying multiple labels.The TTL value in each header is 255, and the third MPLS header has the Bottom of stack flag set to 1, indicating it is the last header in the stack.It's important to consider the increasing overhead for each packet as the network topology and the number of hops increase in SR-TE.

Analyze the flow of packets across links within a stable TE tunnel environment:
In the MPLS-TE scenario Fig3 (A), packets captured on the PE1-P1 link include two types of messages: OSPF hello packets and RSVP SREFRESH messages.Both types of packets are observed in both directions (from 10.1.2.1 and from 10.1.2.2) within the MPLS-TE network .In the SR-TE scenario, the analysis focuses on the packet flow in a stable Traffic Engineering tunnel.Fig3 (B) displays a packet collection captured on the link between PE1 and P1 using WireShark.In this case, only OSPF hello packets are present, and no RSVP messages are observed.Unlike MPLS-TE, SR-TE utilizes the Interior Gateway Protocol (IGP) as the control plane for both IPv4 and IPv6 packets, eliminating the need for additional protocols.In the MPLS-TE scenario, the convergence process was tested by manually shutting down an interface between router P1 and PE1.This caused the tunnel to switch to the E3S Web of Conferences 469, 00075 (2023) ICEGC'2023 https://doi.org/10.1051/e3sconf/202346900075second path choice.By executing the "no shutdown" command on the same interface, updated notifications from IGP OSPF and RSVP were received.In the case of OSPF, LS updates and LS Acknowledge messages were observed.Figure 4 depicts various types of RSVP messages, including Path messages (sent along the path to construct a path state), Resv messages (for creating and maintaining reservation status), and ResvConfirm messages (confirming Resv messages).The convergence procedure resulted in an RSVP overhead of 590 bytes on one link of the Traffic Engineering tunnel, and a similar amount of overhead was observed on other tunnel links.In the SR-TE scenario, the analysis focuses on the convergence process in an SR-TE topology during a network failure.An interface on router P1 was shut down to simulate a failure on the link between P1 and PE1, causing the tunnel to switch to the second path option.The convergence process results, shown in figure 5, display update messages only from the IGP OSPF.LS update and LS Acknowledge messages are observed.The total weight of the four OSPF messages related to convergence is 596 bytes, higher than the OSPF messages shown in Figure 4 (358 bytes).This increase in size is due to the inclusion of OSPFv2 Extended Link Opaque LSA (LSA Type 10) carrying SR-TErelated information.The size of this LSA is proportional to the number of nodes in the path, transmitting more information as the topology grows larger .Opaque LSAs (LSA Type 9, 10, and 11) enhance OSPF functionality, carrying additional information beyond routing information.For MPLS-TE and SR-TE, OSPF carries information such as topology, bandwidth, reserved bandwidth, link coloring, delay, and other attributes.MPLS-TE involves OSPF and RSVP processes during convergence, occupying a total of 948 bytes.In contrast, SR-TE only uses the OSPF process, resulting in a convergence process that occupies less bandwidth (596 bytes).SR-TE utilizes the IGP as the control plane for both IPv4 and IPv6 packets, eliminating the need for additional protocols like RSVP.Therefore, SR-TE achieves a more efficient use of bandwidth during the convergence process.

Conclusion
Both MPLS and SR are powerful technologies that have their own strengths and weaknesses, but when used together, they can optimize network performance.MPLS is a well-established technology with widespread support, while SR offers greater flexibility and control over traffic flows.In this study, two different network topologies, RSVP-TE and SR-TE, were implemented and simulated using EVE-NG in order to analyze their behaviour.The results indicate that SR-TE reduces control plane messages and overhead compared to MPLS-TE.Additionally, SR-TE involves the exchange of OSPF Hello messages, while MPLS-TE requires RSVP and OSPF control plane messages.Despite the overhead in OSPF messages, SR-TE demonstrated a smaller overhead per convergence process.Based on these findings, it is suggested that SR-TE can be advantageous for networks that require faster and more efficient convergence.

Fig 1
Fig 1 Segment Routing Operations

Fig 3
Fig 3 Analyze the packet flow across links within a stable TE tunnel environment (A) MPLS-TE (B) SR-TE Conditions

Fig 3
Fig 3 MPLS-TE convergence process administrators, security professionals, and developers often rely on Wireshark to troubleshoot network issues, analyze network performance, and debug network applications.