Strategies for Improving Failure Protection of 5G Networks

Abstract

This paper researches about reducing the complexity and improving the failure protection of 5G networks by using segment routing in IP infrastructure. Two Scenarios are simulated to prove the resiliency and simplicity of Segment Routing over the currently used WAN signalling protocols LDP and RSVP-TE. Simulation findings show that the Segment Routing fault protection mechanism (TI-LFA) give 100% coverage of the network and simpler to configure and manage. For lower total cost of ownership, multi-vendor approach is preferred for the networks. Interoperability testing for segment routing between different vendor and security hardening of PCE is recommended for future work.

Glossary

Acronym	Meaning
BGP	Border Gateway Protocol
CSPF	Constrained Shortest Path First
DPI	Deep Packet Inspection
ECMP	Equal Cost Multi Path
IGP	Interior Gateway Protocol
LDP	Label Distribution Protocol
LTE	Long Term Evolution
LFA	Loop Free Alternative
MPLS	Multi-Protocol Label Switching
NMS	Network Management Systems
OSPF	Open Shortest Path First
PCE	Path Computation Element
RSVP-TE	Resource ReSerVation Protocol – Traffic Engineering
SDN	Software Defined Networking
SLA	Service Level Agreement
VPN	Virtual Private Network
WAN	Wide Area Network

Introduction

5G is the new technology that is focused not to be used only for mobile voice and data but by variety of end devices with different requirements in terms of latency, bandwidth and reliability. ITU-R envisioned use scenarios for IMT for 2020 and beyond. Figure 1 illustrates use cases of enhanced Mobile Broadband, massive IOT and mission critical IOT services.

Fig. 1. Usage scenarios of IMT for 2020 and beyond

In 5G era, computing and storing of data will be handled by distributed data centres instead of centralized to cope for ultra-reliable and low latency communication. To forward the packets across multiple network domains with special requirements in 5G cannot be fulfilled by the signalling protocols LDP or RSVP-TE.

Segment Routing is the new way of forwarding IP packets over multiple networks without the dependency of signalling protocols like LDP or RSVP which can meet the requirements like scalability, simplification, efficient Traffic Engineering and better fault protection for 5G network.

In Segment Routing to forward the packet from source to destination, path is divided in to segments and each segment has its ID which encoded in to packet header as a stack of one or more MPLS labels. Segment Routing does not need signalling protocols like LDP and RSVP-TE to reserve the network resources or label distribution. It uses extensions of IGP Link State Routing Protocols such as OSPF and ISIS to distribute the labels and link information across the network. There are two ways to deploy segment routing in the network centralised and distributed. In distributed, router that needs to send the packet has to take the decision of path whereas in SDN centralised, path control element (PCE) keeps the information of all the network and takes the decision in selecting the path or reserving the resources which requires in 5G implementation. This helps operator to do the traffic engineering dynamically in a more granular way by using parameter like physical diversity, delay, jitter, bandwidth and number of forwarding hops. Adoption of Segment Routing is relatively easier and prioritized for 5G networks.

LDP and RSVP-TE establish label switch paths for the traffic and keep the path state information on each transit router. 5G networks will bring massive machine to machine diverse communication which in result LDP or RSVP-TE occupy large resources on routers to maintain sessions.

With Segment Routing only the ingress routers keep the path state information. Other routers like egress and transit forward the packets based on the instruction in the form of segments appended to each packet header, which solves 5G scalability problem and can accommodate 5G special treated traffic volume.

Besides scalability, Segment Routing fault protection mechanism TI-LFA provides 100% topology coverage and path optimality.

This paper researches about the Segment Routing Architecture and its uses cases, mainly focuses on research objectives fault protection mechanism TI-LFA and simplicity. Simulation of network failure by using GNS3 assists in comparing the performance of segment routing with LDP and RSVP-TE.

Background

 

In early days when service provider was moving the communication from SDH to IP, Network was dependent upon IGP (Interior Gateway Protocol) to advertise the topology information across the network and which was used to calculate the shortest path. Packet stopped at every hop for lookup and forwarded to the Next Hop in the routing table.

In the 90s, a new way of forwarding packets by using labels MPLS was introduced. It’s growing popularity among Service Providers was due to its flexibility and fast switching.

There are two common protocols used mostly to distribute labels and path selection based on constraints, LDP and RSVP-TE. RSVP-TE is more popular among Service because it supports traffic engineering and better fault protection mechanism. There are several problems faced in deployment and operation of RSVP-TE by Service Providers.

(i)                  Poor load balancing across equal cost paths.

 To overcome this problem, large number of tunnels need to be created in network which cause complexity in management.

(ii)                Every router in the explicit path has to maintain the state of control plane which causes scalability issues in merged networks

(iii)              Operations of RSVP-TE is distributed which makes inefficient decision in allocating resources to the traffic and re-optimization during fault protection.

Networks which handle different type of applications require more flexible, scalable and easy to operate. These requirements cannot be filled by the RSVP-TE.

Segment Routing Architecture

Segment routing forwards the packet along an explicit path with minimal configuration required on nodes. Nodes that can support segment routing, take the segment list as an instruction and forwards the packets according it (Abdullah, Ahmad, & Hussain, 2018). An instruction from segment can be for network or related to any service. Network level instruction directs the node to forwards the packet using shortest path to specific interface. Service level instruction is required to treat the packet differently like deep packet inspection or screening through firewall.

 A segment can be local to segment routing node or has global significance. Segment routing framework depends upon two components:

(i)                  Data Plane

(ii)                Control Plane

Segment Routing Data Plane

This component is responsible for encoding sequence of segments on a packet and forwards the packets according to the instruction. With the segment list there is a pointer as well in header which points the current segment that needs to be processed by the node holding the packet. Each segment has its own segment ID that can be local or global (Filsfils, Nainar, Pignataro, Cardona & Francois, 2015). Local segment ID is locally significant and global must be unique within its domain and is advertised to all nodes part of segment routing. If MPLS is being used as a data plane, then each node maintains a set of local labels reserved for global segment in the SRGB (Segment Routing Global Block). Each node in the domain advertises the global segments as a label range and an index. Each node can be configured different range of SRGB but to make the operation and troubleshooting its recommended to keep the same SRGB block on all the nodes participating in the same segment routing domain. Segment Routing enabled nodes support the following operations

PUSH: This operation inserts a segment at the top of the segment list

NEXT: When the active segment is completed, NEXT instruction moves the pointer to the next segment

CONTINUE: In this operation active segment remains unchanged

There are different types of segments used in segment routing for example IGP segment, BGP peer segment, LDP segment, RSVP-TE LSP segment and BGP LSP segment. In this paper, only the IGP segment will be elaborated. IGP segment represents the attached prefixes or adjacency information of SR capable node which is advertised to other nodes. Different types of IGP segment types are described below.

Fig. 2. Illustration of IGP Prefix/Node Segments.

(i)     IGP Adjacency Segment

Each SR capable router assigns locally significant segment ID for each of its adjacency with other routers. This segment is used when a packet needs to be traversed through a specific adjacency.

(ii)    IGP Prefix Segment

Router participating in segment routing advertises the segments representing its attached prefixes. This is used to forward the packet through shortest path to the egress router where the segment belongs.

(iii)  IGP Node Segment

This segment is assigned to the loopback IP addresses of IGP nodes and their segment IDs are called Node SID. Ingress IGP node forwards traffic to any egress node by pushing the IGP node segment on top of the packet.

(iv)  Service Segment

When special treatment is needed for the traffic flows, service segment with IGP segment forwards the packets through middleboxes like firewall, DPI or Load balancer to perform specific action.

Fig. 3. Sample network for service chaining and traffic engineering use cases.

MPLS and IPv6 are the two data plane technologies supported by SR. Network operators are free to chooses any one of the technologies depends upon their network requirement.

Segment Routing Control Plane

The segment routing control plane has a role of propagating the segment ID information across the network devices. There are two IGP link state protocols OSPF and ISIS which use their extension to advertise segments IDs. Every SR node running ISIS or OSPF keeps the database of all nodes and adjacency segments. Both the IGP protocols have sub-second convergence properties, can quickly update the topology database after the network failure.

There are three methods to compute the SR path:

(i)     Distributed Constraint SPF (CSPF) calculation

In this method, ingress SR nodes calculates the shortest best path to the destination matching the required criteria. It then computes sequence of segments that will be appended to packets.

(ii)    SDN Controller – PCE (Path Computation Element)

In this approach, a centralised computation element is used to calculate the shortest best path and instructs ingress nodes to forward packets by using that path.

(iii)  Statically Configured Path

This method is usually used to test or troubleshoot the scenario by using the statically configured tunnels. Network operators do not prefer this approach because of its scalability, resiliency and management limitation.

Segment Routing Use Cases

Besides simplification of control plane operation, following are some use cases of segment routing.

(i)                  Forwarding of traffic dynamically to the path computed based on required traffic engineering parameters like link congestion, delay or better capacity

(ii)                Enables implementation of service function chaining which performs a special treatment on a packet

(iii)              Centralized operations, administration and management of the network overcomes complexity and human error

(iv)              Effective traffic recovery from failure without requiring signaling on the transit nodes during recovery time

Literature Review

Following related literature and research work are on 5G network requirements and segment routing use cases.

Authors (Choi, Kim, & Park, 2016) presents revolutionary direction for the 5G Mobile Core network to meet the requirements like latency less than 1ms within RAN, data transmission average 1Gbps per user, distributed network architecture to overcome the risk of a single point-of-failure. In this research paper, a reference model and key mechanism is proposed for the implementation of propose direction for 5G Mobile Core Network. The reference model mainly covers RAN and packet core area but lacks IP WAN infrastructure.

(Choi & Park, 2017) emphasizes redesigning of 5G mobile network architecture to meet the most demanding use cases enhanced Mobile Broadband (eMBB), Massive IoT (mIoT), and Mission-critical IoT (cIoT) services. Authors presents a network slice architecture by dividing single physical network into many logical networks with various network functions to support different services independently according to their requirement. The research paper needs further research work on the issue of end to end slice architecture of 5G network.

(Katsalis, Gatzikis & Samdanis, 2018) researches about the need of virtualization in transport network to meet the requirement of 5G mobile communication. Currently VLAN and VPN can isolate the data traffic but may compete for the same physical resources which can not ensure performance in terms of latency, throughput and ECMP. This paper presents SDN based network architecture using Flex-Ethernet as a main candidate interface to slice the transport network with segment routing to allocate the paths dynamically with strict performance guarantees.

(Maila, Marius & Victor, 2017) discuss about the simplification and optimization of IP Networks with Segment Routing. Authors compares segment routing with other label distribution protocols LDP and RSVP-TE in distribution of labels and performance. A logical topology of 6 Cisco Virtual Router instances is used to simulate implementation of segment routing and MPLS LDP separately. The tool used to send packets to the destination is traceroute which output does not show anything about simplification or better fault recovery of segment routing.

(Abdullah, Ahmad, & Hussain, 2018) covers segment routing concept, operation and the working of segment with existing network architectures. Authors researches deeply about SR applications like Traffic engineering techniques, fault protection, network recovery, service function chaining, networking monitoring. Results of the survey shows better performance of SR as compared with the traditional signalling protocols LDP and RSVP-TE in context of link utilization, network resiliency, energy consumption and end-to-end delay.

Dynamic SR operations are presented by (Castoldi, Giorgetti, Sgambelluri, Paolucci & Cugini, 2017) in their research. Authors demonstrated SR dynamic optical bypass and effective load balancing experimentally by using a testbed of virtual controllers and routers. Segment routing provides the efficient traffic engineering with simplified control operations. Experimental validation in this research paper proves segment routing’s efficient resource utilization with simple control operations.

(Giorgetti, Sgambelluri, Paolucci, Cugini & Castoldi, 2017) have focus on two use cases of segment routing: dynamic traffic recovery and traffic engineering in multi-domain network in their research. Both use cases of segment routing simplify the network operations with respect to LDP and RSVP-TE. Authors simulated the network failure scenario using the SR-FAILOVER scheme. Experimental results show segment routing is a strong candidate technology for effective traffic recovery by recovering the network from failure in average of 13 ms and utilizing the resources effectively by creating backup to the destination directly instead of next hop.

(Giorgetti, Sgambelluri, Paolucci & Castoldi, 2015) proposes two procedures for effective traffic recovery (link and node protection) in segment routing network. Authors applied their procedures on multiple networks to find out the number of labels require at the node detecting the failure to direct the traffic on the backup path. Their experimental results do not show up the approximate time for the networks to recover from failure.

Unified MPLS for converged services simplifies the deployment of MPLS in many aspects but still cannot meet the demands of cloud computing. Authors (Cai, Wielosz & Wei, 2014) propose a new evolved architecture that will simplify the network infrastructure with high availability and agility.

 Review Summary

All the literature reviewed in this paper used different methods like survey, simulation, proposal of new architecture or procedures for efficient traffic recovery which makes the segment routing protocol suitable for future networks like 5G Mobile Communication with guaranteed performance and simplified operations. But none of the research paper compares the performance of traditional MPLS protocols LDP or RSVP-TE with segment routing by simulation or running the test scenarios on real networks

In this research paper, experimental method is used to simulate the different network scenarios to compare the segment routing use cases efficient network recovery and simplicity in implementation and control plane operations with LDP and RSVP-TE.

Research Design

Research Objectives

1. How the Segment Routing fault protection mechanism TI-LFA (Topology Independent – Loop-Free Alternative) guarantees the 100% coverage of the Network, unlike IP-Fast Reroute LFA?

 A network topology is simulated on GNS3 tool installed on windows 10 machine and used by both segment routing and IP-FRR one by one. Their performance in terms of coverage and packet loss is compared with each other by using Wireshark and Traffic Generator Ostinato0.9-1.

2. How Segment Routing is better than RSVP-TE in terms of simplicity and fault protection?

During the link or node failure, RSVP-TE Fast-Reroute uses the pre-calculated sub-optimal path as a backup which may increase the latency and link utilization. A link failure scenario is simulated on the network of 4 routers on GNS3 to compare the backup path selection criteria of segment routing with RSVP-TE FRR.

Data Description

Following are the main metrics used to judge the performance

1. Packet loss

• How much number of packets are dropped during the link failure simulated under both segment routing TI-LFA and LDP IP-FRR LFA?

2. Coverage

• Will the Segment Routing TI-LFA able to calculate the backup path at ingress node when LDP LFA fails to do so?

3. Simplicity

• Does the segment routing have the same level of complexity in implementation and control plane operations?

4. Backup Path Selection

• Does the segment routing also steer the traffic on the sub-optimal backup path like RSVP-TE?

Application of Research Methods

Research question 1

 An experiment of simulating network recovery scenario comprised of 4 Cisco CSR1000v instances and traffic generator Ostinato0.9-1 was tested on GNS3. The same logical network topology was configured for segment routing TI-LFA and LDP IP-FRR LFA individually and their performance was compared before and after the network failure. Figure 4 shows network topology used to simulate the traffic recovery scenarios.

Topology Details

• Built the ring topology of 4x virtual instances of Cisco Routers CSR 1000v and Traffic Generator Ostinato0.9-1. 
• Enabled OSPF State Link protocol as IGP (Interior Gateway Protocol) on all routers under Area 0
• For traffic forwarding LDP and segment routing were configured
• Traffic Generator Ostinato0.9-1 connected with CSR1 to generate ICMP traffic to the loopback IP of CSR3

Traffic generator is set to send 1000 packets with rate of 100 packets/sec to the destination Loopback IP of CSR3. In both LDP and segment routing, CSR1 routes the traffic via CSR4 to reach CSR3.

S

D

N

Fig. 4. Topology to simulate traffic recovery scenario

Scenario-1: LDP IP-FRR LFA

Network is configured as LDP to distribute labels across the routers and forward the traffic based on IGP shortest path first algorithm. When the network is steady, CSR1 could pre-install the backup next-hop via CSR2 into the forwarding table path but network topology and its ospf cost does not meet the condition of Loop Free Alternative inequality.

Loop Free Alternative Inequality Condition

D(N,D) < D(N,S) + D(S,D)

10 < 1 + 2

10 < 3

Following output shows the best path to CSR3 from CSR1 without backup path.

Fault Recovery:

To simulate the scenario of traffic recovery after link failure, ICMP traffic is generated by traffic generator Ostinato0.9-1 and link between CSR3 and CSR4 is shut down.

x

Alternate path recalculated

Fig. 6. Topology configured with LDP IP-FRR

1000 ICMP packets are sent by traffic generator to CSR3. When link failure is happened, CSR1 recalculated the alternative path to CSR3 and steer the traffic via CSR2.

Following is the output of traffic generator showing number of frames sent and received. Total number of packets lost are 15.

Scenario-2: Segment Routing TI-LFA

Same network topology is used to configure segment routing. Unlike LDP IP-FRR, segment routing can calculate backup path. Because the pre-installation of backup is not dependent upon either topology. Following output shows the primary and pre-installed backup ready to take traffic instantly if primary path fails.

Wireshark output shows the mpls label 16300 appended into packet header.

Fault Recovery:

To simulate the scenario of traffic recovery after link failure, ICMP traffic was generated by traffic generator Ostinato0.9-1 and link between CSR3 and CSR4 was shut down.

x

Pre-Installed Backup Path

Fig. 7. Topology configured with Segment Routing TI-LFA

1000 ICMP packets are sent by traffic generator to CSR3. When link failure happens, CSR1 recalculates the alternative path to CSR3 and steers the traffic via CSR2.

Following is the output of traffic generator showing number of frames sent and received. Total number of packets lost are 987.

Results and Discussion

Results show that the segment routing TI-LFA gives 100% coverage of network whereas LDP IP-FRR fails to install backup path prior network failure. This pre-installed backup path helps CSR1 to instantly steer the traffic to backup path via CSR2 which results in lesser packet loss as compare to LDP IP-FRR.

 

 

 

 

 

Research question 2

An experiment was conducted to proof that segment routing is better than MPLS RSVP-TE in perspective of simplicity and resource utilization during network failure.

Scenario-1: MPLS RSVP-TE

A RSVP-TE tunnel 13 was established CSR1 ingress to CSR3 egress node. Primary Tunnel 13 chose path via CSR4 as shown in figure 8. To react efficiently during link failure, CSR1 installed backup FRR tunnel which will route the traffic to the next hop via CSR3. It would be a sub-optimal path temporary till the headend CSR1 realizes about the failure.

x

RSVP-TE Primary Tunnel 13

RSVP-TE Backup FRR Tunnel 14

Fig. 8. Topology configured with MPLS RSVP-TE

In normal situation, traffic uses primary tunnel 13 to reach CSR13 via CSR4. Below output shows traceroute to CSR3 loopback IP 192.168.11.3 taken at CSR1.

When link failure happens between CS1 and CSR4, traffic steers the traffic to next-hop via CSR2 then to destination CSR3. There would be inefficient utilization of resources due to that sub-optimal path. Figure shows repetition of hops in using sub-optimal path.

For every destination, traffic engineering tunnel is required. If there are n number of edge nodes, number of tunnels require to configure are 2ⁿ-1. shows tunnels created for primary and backup FRR tunnel.

Every transit router like CSR4 as shown below has to maintain the control plane session for each tunnel passing through it.

Scenario-2: Segment Routing

When segment routing is deployed to forward the traffic from ingress CSR1 to egress CSR3, primary shortest path via CSR4 is preferred. Tunnels are not needed to be configured for egress nodes. Moreover, transit nodes do not keep control plane session of primary or backup paths.

Primary Path

x

Pre-installed Backup Path

Fig. 9. Topology configured with Segment Routing TI-LFA

Below output shows traffic using backup path via CSR2 to reach CSR3 when link failure happens between CSR1 and CSR4. Segment routing calculates the optimal backup path and utilizes the resources efficiently.

Results and Discussion

MPLS RSVP-TE uses suboptimal path as backup FRR to steer traffic to next-hop and then to destination when the protected link fails which results in improper resource utilization. For every edge egress node, traffic tunnel is configured on ingress nodes. As behaviour of TE Tunnels is unidirectional, same number tunnels require from egress to ingress. Complexity in operations and implementation increases with number of edge routers.

On the other hand, segment routing with better backup path selection and simple to configure and manage is getting preferred for the high demand networks like 5G and cloud compute networks.

Limitation and Future Research

Due to the demonstration of traffic recovery scenarios for LDP IP-FRR and segment routing on virtual setup, the difference in packet loss is not very significant due to limited testbed machine memory and software-based decision of forwarding packets by virtual routers. If the same scenario tested on real network, better results might be observed because in high grade routers forwarding plane is handled by hardware.

Segment routing interoperability testing between different vendors routers could not have been conducted due to limitation of license on other vendor virtual router instances. Most of the network operators use multi-vendor equipment to build infrastructure. Before deploying segment routing, many operators may look for proper interoperability test report.

Segment routing deployed in network with centralised approach, can bring security and HA vulnerability to network. If the whole infrastructure is dependent upon only one SDN Controller – PCE for the computation of shortest and best paths, then there would be chances of central point of failure. A centralised model where every NE has individual trust relationship with PCE, needs to be operated with high security. Denial of service attacks on the controller can interrupt the messages to NEs. Future work for the security hardening of the networks intending to use PCE has to be done.

Conclusion

This paper demonstrated the different network scenarios to compare the segment routing use cases efficient network recovery and simplicity in implementation and control plane operations with LDP and RSVP-TE. Results of the first research question prove that segment routing fault protection mechanism TI-LFA provides 100% network coverage and efficient traffic recovery. Findings of the last research question show the simplicity and optimal backup path selection during network failure of segment routing which makes it desirable for next generation networks like 5G.

References

• Abdullah, Z. N., Ahmad, I., & Hussain, I. (2018). Segment Routing in Software Defined Networks: A Survey. IEEE Communications Surveys & Tutorials, 21(1), 464-486.
• Cai, D., Wielosz, A., & Wei, S. (2014, June). Evolve carrier Ethernet architecture with SDN and segment routing. In Proceeding of IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks 2014 (pp. 1-6). IEEE.
• Castoldi, P., Giorgetti, A., Sgambelluri, A., Paolucci, F., & Cugini, F. (2017, July). Segment Routing in multi-layer networks. In 2017 19th International Conference on Transparent Optical Networks (ICTON) (pp. 1-4). IEEE.
• Choi, Y. I., Kim, J. H., & Park, N. I. (2016, October). Revolutionary direction for 5G mobile core network architecture. In 2016 International Conference on Information and Communication Technology Convergence (ICTC) (pp. 992-996). IEEE.
• Choi, Y. I., & Park, N. (2017, July). Slice architecture for 5G core network. In 2017 Ninth international conference on ubiquitous and future networks (ICUFN) (pp. 571-575). IEEE.
• Filsfils, C., Nainar, N. K., Pignataro, C., Cardona, J. C., & Francois, P. (2015, December). The segment routing architecture. In 2015 IEEE Global Communications Conference (GLOBECOM) (pp. 1-6). IEEE.
• Giorgetti, A., Sgambelluri, A., Paolucci, F., Cugini, F., & Castoldi, P. (2017). Segment routing for effective recovery and multi-domain traffic engineering. Journal of Optical Communications and Networking, 9(2), A223-A232.
• Giorgetti, A., Sgambelluri, A., Paolucci, F., & Castoldi, P. (2015, October). Reliable segment routing. In 2015 7th International Workshop on Reliable Networks Design and Modeling (RNDM) (pp. 181-185). IEEE.
• Katsalis, K., Gatzikis, L., & Samdanis, K. (2018, October). Towards Slicing for Transport Networks: The Case of Flex-Ethernet in 5G. In 2018 IEEE Conference on Standards for Communications and Networking (CSCN) (pp. 1-7). IEEE.
• Kos, A. (2016). Segment routing principles and applications for SDN.
• Maila, G., Marius, I., & Victor, C. (2017, March). Segment Routing. In 2017 10th International Symposium on Advanced Topics in Electrical Engineering (ATEE) (pp. 34-38). IEEE.
• Schüller, T., Aschenbruck, N., Chimani, M., Horneffer, M., & Schnitter, S. (2018). Traffic engineering using segment routing and considering requirements of a carrier IP network. IEEE/ACM Transactions on Networking, 26(4), 1851-1864.