4.3. Ring-Based Protection
4.3.1. Introduction/Background
Ring-based protection has been part of the transport network landscape for some time due to the wide deployment of ring network topologies. More recently, interest has arisen in ring-based protection independent of the underlying network topology. In this chapter, the theoretical underpinnings of ring-based protection are reviewed and a wide range of ring implementations and applications are studied. This includes access rings, shared protection transport rings, a variant of the shared protection ring suitable for transoceanic cable use, software defined rings, and a new ringlike protection mechanism known as p-cycles.
4.3.2. Theoretical Background on Rings
Suppose that there are a number of nodes that have to be interconnected. Given that their bandwidth requirements are similar, what is the advantage of a more complex ring topology compared with a simpler (linear) tree topology? The justification given here is a generalization of an example given in [Cahn98]. Certain simplifying assumptions are made in this description, but these do not affect the fundamental results. First, node failures are ignored. Second, it is assumed that all link failures are uncorrelated and that the probability of failure for a link is denoted by p (≤ 1). Taking a global view, the reliability of a network is defined to be the probability that the nodes remain “connected” by working links (i.e., it is possible to reach any node from any other). Given that N nodes are to be interconnected in a network topology, the fewest links are utilized when the topology is a tree. This topology would consist of N-1 links. The probability of failure of a network with a tree topology is the same as the probability that one or more links fail, or conversely, 1 minus the probability that no link fails. This is given by equation (4) below.
Equation 4
If a ring configuration, instead of the tree topology, is chosen, an additional link will be required for a total of N links. The reliability of the ring topology is the same as the probability that two or more links have failed, or conversely, it is 1 – (the probability that no links have failed + that one link has failed). This is given by equation (5) below.
Equation 5
Now, if p << 1, that is, the probability of link failure is small, then using some manipulations based on the binomial expansion of powers of sums and only keeping the dominant terms, the following approximations to the preceding probabilities are obtained:
Equation 6
Equation 7
Equations (6) and (7) confirm the intuitive notions concerning rings and tree (or linear) networks. In particular, rings have a better reliability, that is, the network failure probability decreases as the square of the link failure probability p rather than only at a linear rate for a tree network. Also, the reliability gets worse as the ring size (n) increases. This is because the larger the number of nodes n, the higher the probability that two links will be simultaneously out of service.
4.3.3. UPSRs (Unidirectional Path Switched Rings)
The simplest ring topology to implement is also the one most often used in access networks. A Unidirectional Path-Switched Ring (UPSR) is essentially a 1+1 unidirectional protection scheme incorporated in a ring network architecture. Figure 4-11 illustrates a SONET UPSR. There is a SONET bidirectional fiber transmission pair between each pair of nodes. These pairs form two different “counter-rotating” rings, that is, rings whose transmission directions are in opposite directions. When a node wishes to communicate with another node, it sends two copies of the signal in opposite directions around the ring, that is, on the two different counter-rotating transmission paths. Thus, one of these paths can be considered the working path and the other the protection path.
Figure 4-11. UPSR Example
The “path” part of the UPSR comes from the fact that indications from the SONET STS-Nc path layer (VC-3/VC-4 in SDH) are used to select the better quality signal from the working and protection paths. The “unidirectional” part of a UPSR is a little less obvious since information is being sent along both directions of the ring. Figure 4-11 shows a bidirectional path connection over the UPSR between nodes A and D. The working path from A to D is A-B-D, while the working path from D to A is D-C-A. Hence the two-way traffic between nodes travels only a single direction around the ring under normal circumstances. The advantage of this arrangement is that given a fiber cut anywhere on the ring only one direction of the connection has to be restored (via a very fast select operation).
Note that bidirectional communication between any two nodes requires bandwidth around the entire ring. This may be fine in access situations where all the traffic is heading between “access nodes” and a “head end,” but it is extremely wasteful in more general networking applications.
4.3.4. BLSR–Bidirectional Line Switched Rings (MS Shared Protection Rings)
Bidirectional Line Switched Ring (BLSR) (called Multiplex Section Shared Protection Ring or MS SPRING in SDH) was developed to overcome the bandwidth inefficiency of UPSRs. In the following, we first consider the Two Fiber Bidirectional Line Switched Ring (2F-BLSR) and compare it with UPSR. We then consider Four Fiber BLSR (4F-BLSR). To understand the functionality and advantages of BLSR, service classes in ring networks will be examined, followed by an examination of the operation of BLSRs using K byte protocols.
4.3.4.1. TWO FIBER BLSRs
Figure 4-12 illustrates a 2F-BLSR, where time slots are allocated in each fiber for working and protection traffic. In particular, working traffic on channel m (< N/2) on one fiber is protected by channel N/2+m on the other fiber.
Figure 4-12. 2F-BLSR Network
As seen in Figure 4-12, half the bandwidth in a 2F-BLSR is reserved for protection. Here, there is a connection between nodes A and B and one between nodes A and D. Neither has a preassigned protection connection.
Figure 4-13 illustrates protection in a 2F-BLSR. Here, the link between nodes C and D has failed, disrupting the connection between nodes A and D. Both nodes C and D initiate protection procedures by redirecting traffic the other way around the ring. In addition, they both select traffic that previously came from the failed link from the protection channels. Nodes A and B simply allow the protection channel traffic to pass through. This is one of the reasons why both 2F-BLSRs and 4F-BLSRs are so fast, that is, only two nodes must perform active switching during ring protection. This is quite an advantage over more general mesh protection techniques.
Figure 4-13. Ring Protection in a 2F-BLSR Network
Figure 4-14 illustrates an alternative failure scenario. Comparing this figure to Figure 4-13, it is seen that the same protection channels on the link from NE A to NE C and the link from NE B to NE D are used in both failure scenarios. This is why a BLSR is more bandwidth efficient than a UPSR. SDH therefore refers to BLSRs as shared protection rings.
Figure 4-14. Alternative 2F-BLSR Network Failure Scenario
Considering Figures 4-12 through 4-14, particularly the connection from node A to node D via node C, it can be seen that the time slots used on each link are the same for the connection. This restriction simplifies the amount of information that must be known on the ring to perform a ring switch. This, however, can also exacerbate the bandwidth fragmentation problem described in Chapter 3 (section 3.2.1). An alternative is to allow time slot interchange at the nodes in a ring, that is, each node can, if desired, map the connection to different channels.
4.3.4.2. FOUR FIBER BLSRs
Four fiber BLSR (4F-BLSR) is used when more capacity is needed than what is available in a 2F-BLSR. Each link of a 4F-BLSR contains a dedicated pair of fibers for carrying working traffic and another dedicated pair for protection traffic. Figure 4-15 illustrates the original UPSR network (Figure 4-11), upgraded to a 4F-BLSR. As in a 2F-BLSR, half of the bandwidth on this ring is reserved for protection purposes. A 4F-BLSR, however, has one additional advantage even over a 2F-BLSR due to the fact that both working and protection fiber links are present on each span between nodes on the ring. This is the ability to perform 1:1 span-based protection in addition to ring-based protection.
Figure 4-15. A 4F-BLSR Network
Figure 4-16 depicts the 4F-BLSR protocol for reacting to a break in the working fiber on a single span. Here, a fault has occurred on the working fiber pair between nodes C and D. Because the protection fiber pair is still operational, a simple span switch as in the bidirectional 1:1 linear case is used to restore the working line.
Figure 4-16. Example of Span Protection within a 4F-BLSR Network
The reliability of UPSR and 2F-BLSR topologies are comparable to that of a protected linear network with the same connectivity. Comparing 4F-BLSR with a linear network where every span is protected via linear 1:1 protection, the failure probability p used in equations (3) and (4) (sections 4.2.2 and 4.3.2) is that of the entire 1:1 link failing (not just the working link). Hence, the theoretical reliability of a 4F-BLSR is the best of the structures investigated so far, and its bandwidth efficiency is similar to 1:1 and 2F-BLSR protection mechanisms.
If a span switch will not resolve the failure, a ring switch will be performed as shown in the example in Figure 4-17. Hence, one can think of a 4F-BLSR as actually offering two types of protection: (a) span protection (1:1), of which there can be multiple instances occurring on the ring at the same time, and (b) ring protection of which there can only be one instance on the ring.
Figure 4-17. Example of Ring Protection in a 4F-BLSR Network
4.3.4.3. SERVICE CLASSES IN BLSRs (MS SPRINGs)
Two and four fiber BLSRs support three classes of service: (1) protected traffic, that is, “normal” ring traffic; (2) extra traffic, which uses the protection channels when they are not being used and is subject to preemption; and (3) Non-preemptible Unprotected Traffic (NUT). NUT is traffic that will not be restored in the case of a failure, that is, this traffic will not be “ring switched” (2/4-F BLSR) or “span switched” (4F-BLSR). The traffic, however, will also not be preempted to make way for other traffic. NUT results in bandwidth savings, since it can be carried in both working and protection time slots. Note that this works because the working traffic is unprotected, that is, will never be transferred to the corresponding protection channels. Also, any NUT on the protection channels cannot be preempted.
Besides the bandwidth savings, NUT is a handy service when the transport protocol layers above the SONET STS-Nc (SDH VC-3/4) have their own protection mechanism in operation and interactions between the protection mechanisms at various layers are to be avoided.
NUT does complicate ring operations. For example, ring switches are prevented for the NUT channels all the way around the ring (not just on a span). Hence, information on NUT connections must be distributed around the ring.
4.3.4.4. BLSR OPERATIONS
The virtues of 2F-BLSR over UPSR and those of 4F-BLSR over a 2F-BLSR have been extolled so far. The price paid for the increased efficiency and in the case of 4F-BLSR, the resiliency, is that of a significant increase in complexity. 2/4F-BLSRs are significantly more complicated to develop, implement, and operate than say a 1:N linear protected span. In fact, due to this complexity and due to the incomplete specification of the supporting information, 2/4F-BLSR equipment from different vendors do not typically interoperate. A high level overview of BLSR operations follows.
The most basic information needed concerning a ring (2/4F-BLSR) is the connectivity of the nodes within the ring. For the purposes of utilizing a K1/K2-byte based protection protocol, the number of nodes on a ring is limited to 16. Each node must be numbered, but the numbering, besides the limitation of being in the range 0 to 15, can be arbitrary. A ring map gives the ordering of the nodes around a ring. The ring map for Figure 4-15 would be {A, B, D, C} . It turns out that information concerning the connections active on the ring also needs to be distributed. It was seen that this is needed for NUT, but it is also needed to prevent misconnection of traffic in the case of certain types of failures. The process of preventing misconnection in rings is known as squelching [ITU-T95a]. The distribution of ring maps and connection information around a ring is not standardized as it is typically handled via a management system.
Each node in a BLSR can be in one of three main states: idle, switching, or pass-through. As mentioned earlier, one of the reasons that BLSRs can be so fast is that only two nodes need to be involved in any protection operation. The rest can be either idle (in the case of a span switch somewhere else on the ring) or in the pass-through state (in the case of a ring switch somewhere else on the ring).
K1/K2 byte usage is shown in Table 4-10 and Table 4-11. The bridge request codes of Table 4-10 are given in priority order. In addition, whether these codes are originated internally (int) or externally (ext) is indicated. Note that the destination and source nodes are specified in the K1/K2 bytes.
The precise operation of a BLSR is given in references [ANSI95c] and [ITU-T95a] using a fairly large and elaborate set of rules. The motivated reader is recommended to peruse these specifications for the exact details. To understand the basics of how BLSR works, we may consider the K1 and K2 byte as a “message.” The K2 byte contains the source node ID and the K1 byte contains the destination node ID, and when a failure condition occurs, these messages are sent from the node(s) detecting the failure to the node at the other side of the failure (known from the ring map). Hence, all the nodes in the ring know the identity of the affected link by observing these messages and knowing the ring map. Also there is a bit in the K2 byte indicating whether the message came along the long or the short path around the ring. A node involved in the switch sources these K bytes both ways around the ring. A node in “pass through” state essentially forwards these on with possible changes in some status bits (without changing the command).
| Bits | Bridge Request Code (Bits 1–4) | Ext/Int | Destination Node Identification (Bits 5–8) | |||
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | |||
| 1 | 1 | 1 | 1 | Lockout of Protection (Span) LP-S or Signal Fail (Protection) | Ext | The destination node ID is set to the ID of the node for which the K1 byte is destined. The destination node ID is always that of an adjacent node (except for default APS bytes). |
| 1 | 1 | 1 | 0 | Forced Switch (Span) FS-S | Ext | |
| 1 | 1 | 0 | 1 | Forced Switch (Ring) FS-R | Ext | |
| 1 | 1 | 0 | 0 | Signal Fail (Span) SF-S | Int | |
| 1 | 0 | 1 | 1 | Signal Fail (Ring) SF-R | Int | |
| 1 | 0 | 1 | 0 | Signal Degrade (Protection) SD-P | Int | |
| 1 | 0 | 0 | 1 | Signal Degrade (Span) SD-S | Int | |
| 1 | 0 | 0 | 0 | Signal Degrade (Ring) SD-R | Int | |
| 0 | 1 | 1 | 1 | Manual Switch (Span) MS-S | Ext | |
| 0 | 1 | 1 | 0 | Manual Switch (Ring) MS-R | Ext | |
| 0 | 1 | 0 | 1 | Wait-To-Restore WTR | Int | |
| 0 | 1 | 0 | 0 | Exerciser (Span) EXER-S | Ext | |
| 0 | 0 | 1 | 1 | Exerciser (Ring) EXER-R | Ext | |
| 0 | 0 | 1 | 0 | Reverse Request (Span) RR-S | Int | |
| 0 | 0 | 0 | 1 | Reverse Request (Ring) RR-R | Int | |
| 0 | 0 | 0 | 0 | No Request NR | Int | |
| Note: Reverse Request assumes the priority of the bridge request to which it is the response. | ||||||
| Source Node Identification (Bits 1–4) | Long/Short (Bit 5) | Status (Bits 6–8) | ||
|---|---|---|---|---|
| Source node ID is set to the node's own ID. | 0 | Short path code (S) | 1 1 1 | MS-AIS |
| 1 | Long path code (L) | 1 1 0 | MS-RDI | |
| 1 0 1 | Reserved for future use | |||
| 1 0 0 | Reserved for future use | |||
| 0 1 1 | Extra Traffic on protection channels | |||
| 0 1 0 | Bridged and Switched (Br&Sw) | |||
| 0 0 1 | Bridged (Br) | |||
| 0 0 0 | Idle | |||
4.3.5. Software-Defined BLSRs
It was seen that 4F-BLSR is extremely resilient, that is, given a probability of link failure of a 1:1 link, p << 1, the probability of connection disruption on a 4F-BLSR is on the order of p2. On the other hand, ring network topologies are not the most flexible. Indeed, modern optical cross-connects can be used to construct general topology mesh networks. It therefore might be desirable to use ring-oriented protection schemes in a general mesh network to get the best of both worlds. Figure 4-18 illustrates a mesh network with two software-defined BLSRs protecting a subset of the links. Such “ring emulation” can provide enhanced reliability and increase the restoration speed for connections along certain routes.
Figure 4-18. Two Software-Defined 4F-BLSRs Overlaid on a Mesh Network
4.3.6. Transoceanic Rings
It was stated earlier that BLSR protection is very fast and that part of that speed comes from involving only two nodes in the switching, that is, the two nodes adjacent to the fault. In some situations, this simplified protection action can present a problem. Consider the BLSR network shown in Figure 4-19. A portion of the ring, segments B-E and C-D, cross a large body of water (such as an ocean). A protected connection traversing nodes B-E-F of the ring is shown. Suppose a fault occurs on the ring between nodes E and F. This leads to a BLSR ring switch. With the ordinary BLSR ring switch, it is seen that the circuit, instead of having one transoceanic crossing, now has three! This leads to quite a bit of additional circuit delay and is usually considered unacceptable.
Figure 4-19. Transoceanic BLSR Network Example
The alternative is to route the circuit the other way around the ring, not at the point closest to the fault but at the point where the circuit enters and leaves the ring. This is the essence of the transoceanic variant of the 2/4F-BLSR specified in appendix A of reference [ITU-T95a].
Because more network elements get involved with the recovery of the same line fault now, the time to restore is longer. The target restoration time for a transoceanic ring is 300 mSec, approximately six times longer than that of an ordinary 2/4F-BLSR.
4.3.7. Protection Cycles
Continuing the extension of ring protection from the software-defined 4F-BLSR and the transoceanic ring, a ring-like protection mechanism called the p-cycle is arrived at [Grover+98]. Like the software-defined BLSR, a p-cycle is a ring-like entity that is applied in general mesh networks. Also, like the 2/4F-BLSR, only two network elements are involved in a protection switching operation. As with the transoceanic ring, the switching granularity is at the SONET path (SDH HOVC) level.
Unlike any of the ring mechanisms, however, a p-cycle can protect links that do not lie on the cycle. Figure 4-20 illustrates a network with a p-cycle defined over it. The cycle is the closed loop on the graph A-B-E-F-G-H-D-C-A. Any link on the cycle is protected and an example of this “on cycle” protection is illustrated in Figure 4-21. Connections that used to take the link between nodes A and B will be rerouted over the portion of the p-cycle, B-E-F-G-H-D-C-A.
Figure 4-20. Network Protected by a P-Cycle
Figure 4-21. P-Cycle Protection a Link that Lies on the Cycle
In addition, any link that “straddles” the cycle can also be protected as shown in Figure 4-22. Note how links straddling the cycle actually have two options for restoration. For example, two distinct portions of the cycle C-A-B and B-E-F-G-H-D-C can protect the link between nodes B and C.
Figure 4-22. P-Cycle Protecting a Link that Straddles the Cycle
P-cycles have been investigated for use at the WDM layer [Ellinas+00], SONET path layer [Grover+98], and even the IP layer [Stamatelakis+00]. P-cycles combine the advantages of mesh and ring configurations. With proper placement of connections and cycles over a mesh topology, they can achieve good bandwidth efficiency, that is, similar to that of general mesh. Because only two network elements are involved with a protection switch, their restoration time can approach that of BLSRs. Like a 2F-BLSR, the protection bandwidth is shared by all the links on the ring. In addition, all “straddling” links that are protected also share the p-cycle protection bandwidth. Hence, there is a limit to how large a p-cycle can be. Recall the protocol limit is 16 nodes for 2F-BLSR and 14 lines for 1:N SONET/SDH LAPS. It should be noted that no standardization efforts are yet underway for p-cycle based protection.