Path restoration schemes for MPLS networks VICENTE ALARCON-AQUINO AND MARCELINO MINERO-MUÑOZ

I

t is well known that the Internet is based on a connectionless, unreliable service, which implies no delivery guarantee. It is also recognized that Internet growth has taken an exponential and unstoppable course; at the same time there has been an increasing demand for new and more sophisticated services. Therefore, the technology has had to undergo fundamental changes with respect to the usual practices developed in the mid-1990s. In this super-growth environment, the Internet service providers (ISPs) must find a way to adjust the dramatic growth of network traffic and number of users. To avoid having equipment specifically designed for the new Internet applications, ISPs had to adapt any commercially available equipment. As infrastructure, the asynchronous transfer mode (ATM) switching equipment was the only technology that provided the required bandwidth, packet forwarding capacities, and traffic engineering. The idea was to combine, in many ways, the effectiveness and yield of the ATM switches with the control capabilities of IP routers. The answer was the deployment of the “IP over ATM” model (IP/ATM). The IP/ATM operation supposes the overlapping of a logical topology of IP routers over a physical topology

Digital Object Identifier 10.1109/MPOT.2011.940647 Date of publication: 18 March 2011

22

© EYEWIRE

of ATM switches. The main advantage of IP/ATM is that it executes the translation from a connectionless-based IP data transfer to a connection-oriented-based ATM data transfer. However, the solution that IP/ATM introduces for meeting the cell tax problem is the increase of interconnecting IP nodes. As a result, the Internet Engineering Task Force (IETF) has proposed several service models and mechanisms to provide end-to-end quality of service (QoS). Some of these services are integrated services like resource reservation protocol (RSVP) and multiprotocol label switching (MPLS), among others. The integrated services are characterized by a resource reservation before the data transmission begins, which implies route definition. In these services, the datagram can be assigned to different classes. The MPLS technology was developed after ATM and offers several services, namely, QoS, IP traffic engineering support, and creation of virtual private networks. MPLS inte0278-6648/11/$26.00 © 2011 IEEE

grates layer 2 and layer 3 of the open systems interconnection (OSI) model without discontinuities, and therefore combines the routing control functions of the layer 3 and commutation speed of layer 2 through a network. MPLS technology may be applied to any layer 3 network protocol, although almost all of the interest is in using MPLS with IP traffic. This technology uses signaling protocols to exchange messages between hosts in a network and provides specifications for the routing, forwarding, and switching of traffic flows through a network.

MPLS components MPLS is based on the label switching, also called tag switching, technology that can be characterized by its use of label swapping packet forwarding combined with IP control protocols and a label distribution protocol (LDP). A label is a short fixed-length underlying protocolspecific identifier for a path that a packet IEEE POTENTIALS

should traverse. High-speed switching of data is possible by the insertion of these labels at the beginning of the packet, and this can be used by hardware to switch packets quickly between links. To deploy MPLS in an IP network, the shim header is inserted between the layer 2 and the layer 3 headers. Fig. 1 shows the insertion of the shim header. The label or tag field includes the physical label that identifies the packet. The EXP field is used for QoS implementations. The S field is used to indicate if label stack is present. The time-to-live (TTL) field prevents packets from looping forever in the network. Fig. 2 shows the standard components that form an MPLS network. In MPLS, data transmission through the network occurs on label switched paths (LSPs). LSPs are a sequence of labels at every node along the path from the traffic source to its destination. LSPs are established either prior to data transmission (control-driven) or upon detection of a certain flow of data (data-driven). Label switching routers (LSRs) are the devices that participate in the MPLS protocol mechanisms. LSRs are high-performance routers that provide fast packet forwarding by performing the label swapping process and are located at the core of the network. Label edge routers (LERs), on the other hand, are devices that operate at the edge of the MPLS network. LERs have very important functions in the label distribution process when a packet ingresses into the MPLS network and in the removal of labels when a packet leaves the network to the corresponding access network. The forwarding equivalence class (FEC) is an identifier for a group of packets that share the same requirements for their transport and are forwarded in the same way through an MPLS network. FECs are built with information generated by an interior gateway protocol (IGP), such as open shortest path first (OSPF). In order to support hierarchical routing, MPLS uses the label stack. The label stack is a set of labels attached to a packet organized in a last-in, first-out structure. This stack allows the MPLS network to operate in tunneling mode. This is a unique feature of MPLS for controlling the entire path of a packet without specifying the intermediate routers.

Path restoration schemes When working with real networks, path congestion and link failure are very recurrent problems. It is important for the network manager to support path protection and restoration MARCH/APRIL 2011

THERE ARE SEVERAL SCHEMES THAT ARE USED FOR COMPARISON PURPOSES WHEN A NEW ARCHITECTURE IS PROPOSED. SOME OF THESE SCHEMES ARE HASKIN, MAKAM, AND SIMPLE DYNAMIC AS WELL AS FAST REROUTING RELIABLE FAST REROUTING AND OPTIMAL GUARANTEED ALTERNATE ROUTE.

schemes. These schemes are based on the type of failure, and each one has characteristics that make it preferable over others and depends on the specific requirements of the network. There are several schemes that are used for comparison purposes when a new architecture is proposed. Some of these schemes are Haskin, Makam, and simple dynamic as well as fast rerouting (FR), reliable fast rerouting (RFR), and optimal guaranteed alternate path

Data Link Layer Header

(OGAP). These schemes use an alternate route to forward traffic around a failure that occurs in a primary route and their objectives are to minimize the time of establishment of this alternate route and to avoid the excessive lost of information. The aforementioned schemes are classified as follows: • Local repair: minimizes the amount of time required for failure propagation. Hence, if the restoration can be realized in a local manner it can be accomplished faster. • Global repair: considers that the nodes and links along the primary route are protected by one restoration route. In case of failure, the restoration scheme sends a failure indication signal (FIS) to the ingress LSR node (also known as LER), and when it receives this FIS, the alternate route is activated from this ingress LSR node. It is worth noting that path restoration schemes in reconfigurable hardware for MPLS networks have been proposed by the authors and are reported elsewhere.

Haskin scheme The Haskin path restoration scheme uses alternate routes previously established

Network Layer Header

MPLS Shim Header

Other Layer Headers and Data

32 b

20 b LABEL

3b 1b EXP

S

8b TTL

Fig. 1 MPLS shim header.

MPLS Network

LSR4

R1

LER1

LSR1 IPv4 LER LSR Router

LSR5

LSR6

LER2

LSR2

R2

LSR3 FEC 1 FEC 2 Untagged Packet Tagged Packet LSP1, LSP2

Fig. 2 MPLS components.

23

R4

R5

R6

R7

R0 Failure IPv4 LER LSR Router

R1

R2 R3 Primary Route Alternate Route

Fig. 3 Haskin scheme.

R4

R5

R6

R7

R0 Failure IPv4 LER LSR Router

R1

R2 R3 Primary Route Alternate Route Rout

Fig. 4 Makam scheme.

R4

R5

R6

R7

R0 Failure IPv4 LER LSR Router

R1

R2 R3 Primary Route Alternate Route

Fig. 5 Simple dynamic scheme.

with local repair (see Fig. 3). In this scheme the network topology allows the establishment of the alternate route between the ingress and egress LSRs (also known as LERs) of the LSP tunnel in such way that the alternate LSP does not share any resource with the route to be protected. The main idea of this scheme is to return the traffic from the point of failure on the protected LSP to the ingress LSR so that the traffic could be redirected through an alternate route between the ingress LSR and the egress LSR of the protected tunnel. The alternate route is established as follows: • The initial segment of the alternate LSP is between the last hop LSR 24

THE FAST REROUTING SCHEME ADDRESSES THE DRAWBACKS OF THE HASKIN SCHEME WITH RESPECT TO ROUND-TRIP DELAY AND PACKET DISORDER DURING RESTORATION. THESE ASPECTS ARE ADDRESSED BY A SWITCHOVER PROCEDURE INITIATED IN THE LSR THAT DETECTS A FAILURE.

before the point of failure (R2) and the ingress LSR or LER (R0) in the opposite direction of the protected LSP. • The final segment of the alternate route is defined between the ingress LSR (R0) and the egress LSR (R7), that is, R0R4-R5-R6-R7.

Makam scheme The Makam path restoration scheme uses global repair and allows dynamic and prenegotiated activation of the alternate route (see Fig. 4). However, the dynamically-established alternate routes add more time to the restoration operation compared with the prenegotiated activation.

The establishment of the alternate route for this scheme is as follows: • When a failure is detected, the node detecting the failure (R2) sends an FIS to the ingress LSR node (R0). • All packets in transit between the failure detection and the moment in which the FIS arrive to the ingress LSR node are lost. • When the ingress LSR node (R0) receives the FIS redirects the traffic through an alternate route to the egress LSR node (R7). The main difference between this scheme and the Haskin scheme is that it does not redirect the traffic from the point of failure. Instead, it redirects the traffic from the ingress LSR node (R0).

Simple dynamic scheme The simple dynamic scheme uses local repair and dynamic activation. Hence, the alternate route is established when the point of failure is detected (see Fig. 5). When a failure in the primary route occurs (R0-R1-R2-R3-R7), the node that detects the failure (R2) establishes a backup path or alternate route (R0-R1R2-R5-R6-R7) by rerouting the traffic flow through the nearest usable link toward the destination node. This scheme can consider link failures as well.

Fast rerouting scheme The FR scheme addresses the drawbacks of the Haskin scheme with respect to round-trip delay and packet disorder during restoration. These aspects are addressed by a switchover procedure initiated in the LSR that detects a failure (see Fig. 6). This switchover procedure consists of sending the packets from the LSR that detect the failure (R2) backward through the primary LSP (R1-R0). This switchover IEEE POTENTIALS

R4

R5

R6

18 17 Hold On Packets

R0 5

R1

6

7

R2

15

LER

LSR

R7

R3 16

Failure 10

11 12 13 14

IPv4 Router

9

8

4

3

2

1

Lost Packets

Buffer R1

Switchover Procedure

Primary Route

Tagged Packet

Alternate Route

Failure Indication Signal

Fig. 6 Fast rerouting scheme.

LSP to the stored packets (packet 11 to 14 in R1). According to Fig. 6 the alternate LSP is formed by the following elements: R1-R0-R4-R5-R6-R7.

procedure does not avoid the lost packets in transit between the node that detects the failure and the node with the failure (packets 1 through 4). When the nodes in the backward LSP detect these packets, they start storing the incoming packets in a local buffer (buffer R1). The last packet forwarded to the primary LSP by these nodes before initiating storage is tagged (packet 10) in order to be identified when it is on its way back. By doing this, the order of the packets is preserved since the intermediate nodes (R1 and R0) in the backward LSP examine the tagged packet and determine when to start sending through the backward

RFR, each LSR in the primary route has a buffer into which a copy of each packet is stored while it is forwarded through the primary route. In this scheme the packets in transit between R2 and the failure in R3 are stored in the buffer of R2.

Reliable and fast rerouting scheme The factors that affect the FR scheme are packet loss, traffic recovery delay, and packet disorder. The RFR scheme eliminates both packet loss and packet disorder during the restoration period (see Fig. 7). In this scheme, as soon as an LSR node in the primary route (protected route) detects a failure, a switchover is established and packets are tagged and sent back through the backward LSP. In

R4

R5

Optimal and guaranteed alternate path scheme When a failure exists in a primary LSP, the backup LSP is used and thus this becomes a new primary LSP, which is not protected by an alternate LSP. This problem is addressed by the OGAP scheme. The OGAP scheme avoids the problems created when there is a change

R6

18 17 Hold On Packets

R0 1

16

R1

1

4

R3

15

Failure 10

16 Buffer R0

LER

LSR

1 IPv4 Router

R7

R2

11

10 16 Buffer R1

5

4 1 5 2 3 4 10 Buffer R2

3

2

1 Switchover Procedure Primary Route

Tagged Packet

Alternate Route

Failure Indication Signal

Fig. 7 Reliable and fast rerouting scheme.

MARCH/APRIL 2011

25

R8

R7

R4

R5

R6

18 17 Hold On Packets

R0 1 16

R1

1

R3 4

R2

15

Failure 10

16 Buffer R0

LER

LSR

IPv4 Router

1

11

10

16

5

4 5 1 2 3 4 10 Buffer R2

Buffer R1

3

2 1

Primary Route

Switchover Procedure

New Primary Route (NPR)

Tagged Packet

Backup Route for NPR

Failure Indication Signal

Fig. 8 Optimal and guaranteed alternative path scheme.

in the backup LSP after it is calculated. The search for a backup LSP for the new primary LSP is accomplished when the traffic is being conveyed through the new primary LSP. The OGAP characteristics are a low packet disorder during restoration and the ability to handle several failures in a network, not just a single failure. According to Fig. 8, the primary route is formed by the following elements: R0-R1-R2-R3-R9. The new primary route (NPR) for the failure shown is formed by R1-R0-R4-R5-R6-R9, and a possible backup route for the NPR may be formed by R4-R0-R7-R8-R9.

OMNET11 IS AN OBJECT-ORIENTED OPEN SOURCE SOFTWARE USED TO SIMULATE, IN MODULAR FORM, DISCRETE EVENTS AND HAS A GRAPHICAL USER INTERFACE USED TO DEVELOP SIMULATIONS OF DIFFERENT COMMUNICATION NETWORKS.

Simulation results To assess the performance of the aforementioned path restoration schemes simulations in terms of arrival time to destination are carried out in OMNET11. OMNET11 is an objectoriented open source software used to simulate, in modular form, discrete events and has a graphical user interface used to develop simulations of different communication networks. It also uses a network description (NED) language for the graphical construction of the network. OMNET++ simulations can be carried out in parallel form providing results similar to those obtained in real networks. To perform the simulations we use the network topology shown in Fig. 9. Simulations for all models are carried out 26

by sending a trivial file transfer protocol (TFTP) packet from host H2 to host H3 of 28 kb in 19 fragments in order to guarantee the payload in the data link layer. This network consists of four IPv4 routers (R1, R2, R3, and R4) connected to an MPLS network that consists of 10 LSRs and four LERs (LER1, LER2, LER3, LER4). For each one of the simulated schemes, the primary route (working path) consists of the nodes R2-LER2LSR2-LSR5-LSR8-LER3-R3. In addition, for each one of the path restoration schemes a failure in the link between nodes LSR5 and LSR8 is simulated after seven fragments of the TFTP message are sent toward their destination. The remaining fragments are sent to the destination using an alternate route accord-

ing to the operation rules of each scheme. In the case of the Haskin, Makam, FR, RFR, and OGAP schemes the alternate routes (backup path) are formed by the following elements: LER2LSR1-LSR4-LSR7-LSR10-LER3. In the case of the simple dynamic scheme, the alternate route (backup path) is formed by LER2-LSR2-LSR5-LSR7-LSR10-LER3. The simulation results show that the simple dynamic scheme has the lowest arrival time when a failure in the primary route occurs (see Table 1). Furthermore, the time required to establish the alternate route is minimal, and so the packet loss is reduced when compared to Haskin and Makam schemes. Note that the label distribution setup in the analyzed MPLS network takes 170 ms for all the simulated schemes. From the results obtained we can also observe that the transit time used to send a packet from origin to destination for Haskin and Makam schemes along with packet loss is a consequence of sending back packets from the point of failure to the ingress LSR (LER2). In addition, FR and RFR schemes provide similar transit times when compared to the Haskin scheme since these schemes are Haskin-based. The only difference is that they require additional buffers in the LSR that constitute the primary LSP. As a result, these buffers involve an additional memory time processing (MTP). For simulation purposes, the buffers (with size 8) IEEE POTENTIALS

H4

200.52.207.48/28 Net_02

H3

223.27.9.0/24 MPLS_Net

LER4

included in the primary LSP add a time of 23.55 microseconds per data link layer packet in terms of MTP for a CPU with a clock frequency of 50 MHz. Moreover, the OGAP scheme provides an increased arrival time since it requires an additional time of 110 ms to establish the NPR. In this paper, the OGAP simulation is carried out without considering the activation of the NPR.

LSR9

LSR10 LSR4 LSR1

Backup Path R1 H1

Fig. 9 Simulated network for the Haskin scheme.

LER1

LSR2

LSR3

In:64 Out:65 In:63 Out:64

Conclusion

Working Path

LSR6

LSR7

In:65 Out:66

LSR8 LSR5

Failure LER2

200.57.157.160/28 Net_01

H2

R2

Out:63

In:66 Out:67

In:0

LER3

R4

R3 2) LDP Session Establishment • Determine Active/Passive Role • Send/Receive LDP Initialization Message 4) Label Switched Path Created • Send/Receive LDP Keep Alive Message LDP Process In:61 Out:62 In:374580 Out:304153 In:304153 Out:0

3) LDP Distribution and Management • Send LDP Address Message to Peers • Send LDP Request Message • Send LDP Mapping Message 1) LDP Basic Discovery Mechanism • Sens/Receive LDP Hello Message Haskin_Compare

MARCH/APRIL 2011

THE SIMULATION RESULTS SHOW THAT THE SIMPLE DYNAMIC SCHEME HAS THE LOWEST ARRIVAL TIME WHEN A FAILURE IN THE PRIMARY ROUTE OCCURS.

We have presented an overview and simulation of several path restoration schemes for MPLS networks. Each scheme has its own advantages and disadvantages that make them appropriate for different applications. These characteristics include recovery time in case of failure, packet loss, packet reordering, and recovery delay. The decision of implementing some of these schemes in a real network depends on how much resources are available and the requirements of the network user’s. Furthermore, simulation results based on OMNET11 show that the simple dynamic scheme has the lowest arrival time when compared to the other reported path restoration schemes. Future work will be focused on investigating further comparisons and enhancements of these schemes.

Read more about it • D. Haskin and R. Krishnan. (2000, May). A method for setting an alternative label switched path to handle fast reroute [Online]. Available: http:// tools.ietf.org/html/draft-haskin-mplsfast-reroute-04 • S. Makam, V. Sharma, K. Owens, and C. Huang. (1999, Oct.). Protection/restoration of MPLS networks [Online]. Available : http ://tools.ietf. org / ht m l /draft-makam-mpls-protection-00 • G. Ahn and W. Chun, “Design and implementation of MPLS network 27

received Ph.D. and D.I.C. degrees in performance moniTransit Time Memory toring of communication Path Message NPR from Time Arrival networks using wavelet Restoration Time Time Origin to Processing Time to Scheme Departure Creation Destination (MTP) Destination transforms from Imperial College London in the UnitHaskin 170 ms – 130 ms – 300 ms ed Kingdom. His current Makam 170 ms – 150 ms – 320 ms research interests include Simple Dynamic 170 ms – 90 ms – 260 ms security in communication Fast Rerouting 170 ms – 130 ms 0.376 ms 300.376 ms networks, wavelet theory (FR) applied to performance Reliable FR 170 ms – 130 ms 0.753 ms 300.753 ms moni toring of communiOGAP 170 ms 110 ms 130 ms 0.753 ms 410.753 ms cation networks, waveletbased image processing, • A. Varga. (2001). OsMNeT++ dissimulator (MNS) supporting QoS,” in Proc. and path restoration in MPLS networks. crete event simulation system [Online]. 15th IEEE Int. Conf. Information NetworkMarcelino Minero-Muñoz (marcelino. Available: https://labo4g.enstb.fr/twiki/ ing (ICOIN’01), pp. 694–699, Feb. 2001. [email protected]) received the M.Sc. pub/Simulator/SimulatorReferences/ • L. H. Gonfa, “Enhanced fast redegree in electronics from the Univeresm2001-meth48.pdf routing mechanisms for protected traffic sidad de las Americas Puebla (UDLAP), in MPLS networks,” Ph.D. dissertation, Mexico. He is currently pursuing his Universitat Politécnica de Catalunya, Ph.D. degree at Imperial College LonAbout the authors Apr. 2003. don in the United Kingdom, in the area Vicente Alarcon-Aquino (vicente.alarof intelligent systems and networks. He [email protected]) is a full-time professor in • M. Minero-Muñoz and V. Alarconwas previously a full-time professor at the Department of Computing, ElectronAquino, “Reconfigurable path restoration UDLAP in the Department of Computics, and Mechatronics at the Universischemes for MPLS networks,” J. Comput. ing, Electronics, and Mechatronics. dad de las Americas Puebla, Mexico. He Sci., vol. 8, no. 2, pp. 29–38, June 2009. Table 1. Arrival times for simulated path restoration schemes.

28

IEEE POTENTIALS