Characterization of the Spanning Tree Protocol

Transcription

1 Characterization of the Spanning Tree Protocol Eduard Bonada Master thesis submitted in partial fulfillment of the requirements for the degree: Master Program in Information, Communication and Audiovisual Media Technologies Supervisor: Dolors Sala Department of Information and Communication Technologies Universitat Pompeu Fabra Spain September 7

2 To me, to her, to them

3 Contents Acknowledgements v Abstract Introduction. Bridged networks Loop avoidance techniques About this work Spanning Tree Protocol. Overview Tree construction Protocol notation BPDU description Protocol operation BPDU reception Reconfiguration procedure Timeout of stored information Pseudo-code Initial convergence example Root Bridge Selection Root Port Selection Designated Port Selection Global example Topology activation Port States Reconfiguration due to failures Link failure Root Bridge failure Topology change mechanism Convergence time Initial convergence Conclusion on the initial convergence Convergence after a failure Conclusion on the convergence after a failure iii

4 3 STP timers analysis 3 3. Parameters glossary Timeout of the old topology: the message age timer Message Age Timer expression Reaching all the network Effect of constant delays Modeling Variable contributions Lost BPDUs Final expression Mapping the IEEE standard Parameter adjustment Choosing the diameter Adjustment of the estimated bridge transit delay Activation of the new topology: the forwarding delay timer Forwarding Delay Timer expression Worst case of maximum distanced timeouts Mapping the IEEE standard STP extensions Rapid Spanning Tree Protocol Port states and roles Immediate transition to Forwarding State Keep-alive BPDUs No Timers: Proposal/Agreement handshake Topology Change Notification Other proposals Shortest Path Spanning Trees Link state routing approaches Conclusions and future work 6 References 65 iv

5 Acknowledgements I especially would like to thank Dr. Dolors Sala for her guidance and support as well as for the patient work on reviewing the final document. I also wish to thank my colleagues for their comments on many different aspects that have helped to the completion of the current work. I dedicate them a little part of the current document. Adrià and his knowledge on Latex (equations in section 3..) Cristina and their comments on simulation (pseudo-code in section..3.4) Carles and his motivating energy (Fig..6) Anna and everything I needed from her (the bibliography). v

6 Abstract The simplicity, cost-effectiveness and economies of scale make Ethernet a popular choice for local area networks (LANs) solution. These features together with the fact that more than 9% of IP traffic is generated in Ethernet LANs makes this technology a good candidate for an end-to-end solution across larger networks. However, this new application of Ethernet demands for a set of new requirements. For example, in a service provider network a mandatory aspect is to perform with the maximum efficiency and providing high availability. Since initial protocols were designed to be used in small networks, in this new environment they do not meet these requirements. The Spanning Tree Protocol (STP) is the responsible of building the active tree-shaped topology running on top of the physical network. It takes the order of tens of seconds to create the tree, and this is totally impossible to be imagined for provider networks, which specify a maximum bound of 5msec. Some improvements have been proposed and implemented, but a global solution that achieves all requirements has not been designed yet. This Master Thesis aims at deeply study and characterize the operation of the STP. Since this protocol is the basis of all the versions and enhancements published so far, it is important to start studying the foundations.

7 Chapter Introduction The Broadband access networks research area is growing according to the home multimedia services needs. Since Internet has become as popular as it is, and the end user starts needing even more and more services running over the network, the research and development of new and better broadband technologies are required. For instance, in the early future, or even nowadays but not a very extensive way, Triple Play services (data, video, voice) will become ubiquitous. One of the most important elements taking part in this development is the Ethernet technology. Nowadays, it has been consolidated as the predominant standard in Local Area Networks (LAN) in front of other technologies such FDDI, ATM or SDH. This has been achieved thanks to its high features, compatibility with interfaces working at different rates, economy, auto-configuration capacity and independency from IP addressing. In fact, Ethernet capacity has grown more than times in 3 years, and initial switching devices, able to connect only a few networks, have become N*GB switches, close to the popular Moore law. Besides, Fast and Gigabit Ethernet bring more bandwidth to the technology. Moreover, It has been estimated that more than 9 percent of IP traffic originates from Ethernet LANs []. Making Ethernet an end-to-end technology spanning across LANs, metropolitan area networks (MANs), and possibly wide area networks (WANs) is starting to be a reality. Moreover, in the last years there has been an important growing in the interest of using Ethernet as a service provider. Leading this action there is Metro Ethernet Forum [], an industry organization dedicated to accelerating the worldwide adoption of Ethernet for managed services. Another aspect that mainly relates this Ethernet evolution is the change from using the shared medium, originally in Ethernet, to a point-to-point framework. Note that dedicated links are necessary to obtain the best performance and simplify the link protocols, and shared links are much less efficient due to collisions. Since Ethernet LANs are mainly point-to-point networks and bear with large distances, this evolution makes both LANs and WANs to converge to the same technology solution.

8 . Bridged networks A single Local Area Network (LAN) permits the connection of a limited number of hosts over a limited physical extension. Interconnecting a larger number of hosts that are not in close proximity usually requires the use of multiple of such LANs and a solution to allow communication between hosts on different networks. Bridges are devices that provide this interconnection forming an extended or bridged LAN [3]. They operate on top of Medium Access Control (MAC) layer, which actually is a sublayer of the data link layer. The data unit in this layer is the MAC frame. Hardwired MAC addresses are used to identify hosts and different bridges interfaces (connecting different LANs). These are the bridge operations: Frame Forwarding. The basic function is to forward MAC frames from one LAN to another. This operation is based on the forwarding tables that the bridges build themselves (Address Learning). Bridges do not modify the content or format of the frames they receive, these are just forwarded. Since no content is modified, a primary advantages of bridging is that it remains totally transparent to upper-layer protocols. This behavior is contrary to routers operation and allows a faster frame forwarding. Address Learning. Initially, a bridge has neither knowledge of host location nor routes to get it. The device must learn from frames coming into its ports to figure out through which segment each host can be reached. When a bridge receives a frame from S in port P, it assumes that in order to send a frame to S, it must be forwarded to port P. As learning process continues, bridges build the forwarding tables that map the MAC addresses of the end hosts with the outgoing ports of the current bridge. Broadcast. When a bridges receives a frame with the broadcast destination, it sends it through all the ports except the receiving one. This In this way broadcast messages can reach all available L network, which is one of the basic features of the link layer technologies like Ethernet. Unicast Flooding. If a frames arrives at a bridge and its destination is not found in the forwarding table (not yet learned), the bridge cannot determine the outgoing port. In this case the bridge treats the frame as it were a broadcast and sends it out through all remaining ports. One of the results of using a bridged LAN is that path redundancy occurs. On one hand this is a good point which the network performance can take advantage of. In case of possible device failures, alternative paths traversing other bridges can be used. On the other one, the fact of having redundancy ends in the possible existence of links forming physical loops. This must absolutely be avoided because otherwise the basic bridge operations (basically the broadcast) make the network totally unstable [4]: Broadcast storm. Without a loop avoidance mechanism, each bridge will endlessly flood broadcast packets to all ports. Fig.. shows an example of this situation. When Host sends a broadcast frame, like an 3

9 ARP request to Router, the frame is received by bridge A, who identifies it as broadcast data and floods it into segment B. When the broadcast frame arrives at bridge B, this will repeat the previous process flooding it to Segment A. Thus, the frame will endlessly travel around the loop network even Router will have already received it many times. Figure.: Broadcast Storm in Redundant Networks without a loop-avoidance technique Filtering Database Instability. When multiple copies of a frame arrive at different ports of a bridge, a bad MAC address learning occur. In Fig.. the Host sends a unicast frame to the Router (source MAC address is the Hosts MAC and the destination MAC address is Routers MAC). Both bridge A and B will receive this frame and learn MAC address of the Host on Port. However, as bridge A has not yet learned the MAC address of Router it will flood a copy of the received frame to segment B. When the copy of the frame from bridge A arrives at bridge B, this will remove the first entry (Host MAC address on Port ) in Filtering Database and add a new mapping of Host MAC address on Port. Bridge B incorrectly learns Host MAC address on Port. Therefore, bridge B can not forward frames properly because the instability of mapping MAC address to each port. Figure.: Forwarding Database instability in Redundant Networks without a loop-avoidance technique 4

10 . Loop avoidance techniques The correct functioning of the basic bridge operation requires that the network must have only one path between any pair of hosts [3] [5]. In other words, if the active topology is an spanning tree, the restriction is solved. The first solution to this problem is first presented in the seminal paper of [6]. It describes a distributed algorithm that creates a tree-shaped topology on top of the physical network and spanning from a bridge elected as the root. Since the active topology is tree-shaped, no loops exist, and the broadcast operation can be safely executed. Basically, the protocol breaks the loops by blocking some bridge ports. The protocol serves as a basis for the IEEE 8.D workgroup and the standardized version first appeared in 99 and a final revision in 998 [7], under the name of Spanning Tree Protocol (STP). The STP protocol is originally designed to be run in LANs with not very restrictive Quality of Service (QoS). For example without requiring a fast reconfiguration or a high-efficient use of bandwidth [8]. In terms of load balancing, STP is quite ineffective. It bases the loop-breaking technique in avoiding the use of some of the network links (blocking ports). This automatically results in an low efficiency of the network possibilities. Besides, a great amount of traffic passes through the root bridge and high congestion occur in root proximities. As the new application of Ethernet bridged networks resides in large networks, this inconvenience becomes even more impracticable. Another of the main drawbacks when using STP is large networks is the reconfiguration time after a change in the topology (i.e a failure). It takes the order of 3-5 seconds until the new tree is active again and data forwarding can be totally restored. The reason is because it is a extremely conservative protocol based on large timers to avoid any problem (loops) in the network. If the technology is wanted to be used in provider networks, a maximum reconfiguration time of 5 ms must be reached. This issue is an important point in the STP analysis, and the characterization of the protocol operation aims at describing the reasons why a reconfiguration takes the order of one minute. Therefore, in large networks requiring high-availability the use of the original STP is not possible. Consequently, the same IEEE 8.D workgroup presents a new version of the standard including an improvement of the STP now providing a rapid reconfiguration: the Rapid Spanning Tree Protocol [9]. This revision of the original STP includes changes mainly related to rapid transitions removing timers. So far, many alternate approaches have been published in order to correctly adapt the STP to this new environment, but a global successful solution has not been achieved yet. Today, most of the research is focused on trying to provide extra functionalities to the current protocols. For example an important research topic is the use of multiple spanning trees on the same network. This technique aims at using all links in the network and hence improve the bandwidth efficiency. The IEEE specification describes how to apply VLANs to deal with this issue[]. Other publications are more focused providing QoS and Traffic Engineering [] [] [3]. Another approach also using multiple instances of spanning trees points at always use the shortest path between any pair of nodes [5] []. Note that the original spanning tree paths follow the shortest path only from the root to other nodes, but not to different nodes of different branches. 5

11 .3 About this work In any of the previous cases the approaches use the philosophy of the STP when creating each one of these multiple trees. Then it is worth noting that the basis of everything is the original protocol. Even though the multiple tree solutions propose to use the RSTP, this is at the same time based on the first protocol. Consequently, we have the feeling that the key is locating the starting point in the study of the basic STP. This Master Thesis focuses on the description of the protocol operation in order to exactly understand how it works and which are the main properties of its behavior to refine or redefine the protocol. Therefore, the main objective of this work is to characterize the properties of the STP. In order to achieve the objective, the following methodology steps are followed: Documentation review. The IEEE standard [7] and only a few articles are available in published literature. Simulator implementation. This step aims at getting a comprehensive understanding of the protocol operation. Formal analysis of the protocol behavior. Once the protocol functioning is clear, simple mathematical equations are derived. From a more general point of view, this task is only the first part of a bigger work. This Master thesis is the first milestone of a line of research on logical topology protocols in the IEEE 8. architecture. This would include a characterization of the RSTP at the same level, reviews of other techniques, and finally the provision of a new proposal. However and because of time requirements, this is rather a work for a PhD thesis. This document is organized as follows. Chapter provides a detailed description of the protocol operation. It presents the notation, elements involved and some examples for a better understanding of the protocol functioning. Afterwards, chapter 3 contains an analysis of the STP timers. Chapter 4 introduces the extensions that have been developed in order to improve the original STP. Finally, chapter 5 concludes the work with the evaluation of the results and the future work.

12 Chapter Spanning Tree Protocol The fisrt step on the path towards a spanning tree active topology was done during the golden period of the bridged networks by Radia Perlman [6]. Perlman is considered the inventor of the Spanning Tree Algorithm that creates an active tree-shaped topology, hence avoiding loops, on top of the physical network. The protocol was later adopted by the IEEE 8.D standard on its versions of 99 and 998 [7] (in the last version of 4 [9] it has been replaced by the RSTP). Both the original Perlman s description and the IEEE version follow the same philosophy of the algorithm, but this chapter deals with the description of the IEEE 8.D version.. Overview The Spanning Tree Protocol (STP) is a link management protocol that provides path redundancy while preventing undesirable loops in the network. For Layer Bridged Networks to function properly, only one active path can exist between two stations. The most common reason you find loops in networks is the result of a deliberate attempt to provide redundancy: in case one link or bridge fails, another link or bridge can take over. However, multiple active paths between stations cause loops in the network. STP is a technology that allows bridges to communicate with each other to discover physical loops in the network. The protocol specifies an algorithm that bridges can use to create a loop-free logical topology. In other words, STP creates a tree structure of loop-free leaves that spans the entire bridged network. STP operation is transparent to end stations, which are unaware whether they are connected to a single LAN segment or a bridgeed LAN of multiple segments. In order to break the loops, bridges talk to each other and take decisions about keeping ports active or blocking them following the protocol rules. Network bridges exchange information containing their state by transmitting special frames called Bridge Protocol Data Units (BPDUs). The STP uses the BPDU information to elect the Root bridge of the bridged network, as well as the root and designated port for each segment. This chapter aims at describing the elements and operation of the STP towards the building of the tree-shaped active topology. To do so, section. first explains the basic algorithm and the main protocol elements. Then, section

13 .3 describes how the protocol really activates the ports belonging to the tree and leaves blocked the rest. Finally, the reconfiguration process due to failures is addressed in section.4.. Tree construction This first section aims at describing how the protocol operates in order to create the tree topology. STP is a distributed protocol, therefore each bridge proceeds individually. Eventually, all singles bridges agree on a topology, which is treeshaped and hence loop-free... Protocol notation The STP creates a connected active topology from the components of a Bridged Local Area Network. In an stable state, after initialization or a change of topology, the ports of the bridges can be activated or not. Only ones in a called Forwarding mode can receive and transmit data frames, while others, in a blocking mode, can not. This is called the active topology of the tree, where all ports take part (including those in the blocking state because they may become active if a topology change occurs). In this tree, there is one bridge which is known as the Root Bridge of the network. It performs as the root of the tree and other bridges are connected to it through different network segments. These are called the Designated Bridges and forward frames from their Designated Ports to the Root Port towards the Root Bridge. Fig.. pictures an example. R D D D ROOT BRIDGE R Root Port D Designated Port A Alternate Port Forwarding tree D R A R A 3 D Figure.: STP notation Each bridge must know at every moment which is its position within the tree topology. This is done by assigning different roles to the bridge ports: the Root port is the closest port to the Root bridge in terms of path cost. The election of the root port is based on the comparison of port priority vectors that are carried inside a BPDU and stored in the local port database. 3

14 A port Designated Port is the one with the best priority vector on the LAN where it is connected. All bridges connected to a LAN segment listen others BPDUs and agree in which bridge will act as Designated. There is only one Designated port in each network segment, and this is the responsible of forwarding frames from that segment. Alternate ports are inactive ports in the tree but aware of possible topology updates to enter in activity when needed. Note that any change in the active topology may result in a change on the ports roles, since a port that was Alternate may become Designated or Root, and vice-versa. In order to really break the loops, all Alternate ports are put into an state called blocking. Other ports will remain in an active state, called Forwarding, until a new reconfigurations may change the port states (refer to section.3 for more information)... BPDU description The correct operation of the protocol relies on the exchange of information between neighbor bridges. This is one of the basis of the STP since all the decisions that the algorithm takes are based on comparisons between current bridge state and received information. To do so, BPDU are exchanged between neighbors and information is propagated to all the elements. These control frames contain information regarding the sending bridge and will be used in the receiver to take protocol decisions, if necessary. There exist two types of BPDUs: Configuration BPDU: used for the computation of spanning tree Topology Change BPDU: used to announce changes in the network topology (described in section.4) Fig.. shows the fields of a Configuration BPDU. Among other parameters, the most relevant fields to understand the tree construction are: Bridge Identifier. The identifier of the transmitting Bridge. Root Identifier. The unique identifier of the Bridge that the transmitter believes to be the Root. Root Path Cost. The cost of the path to the Root from the transmitting Port. Port Identifier. The identifier of the transmitting Port. Each bridge has an identifier. It is unique and it must be possible tune its value by manual configuration. To do so, the Bridge Identifier and Root Identifier are values consisting of the following fields: Bridge Priority. The priority or weight of a bridge in relation to all other bridges. The priority field can have a value of to 65,535 and defaults to 3,768. MAC Address. This is the MAC physical address, which is hardcoded and unique for each network interface. 4

15 Protocol Identifier Protocol Version Identifier BPDU Type Flags Octet Root Identifier Root Path Cost Bridge Identifier Port Identifier Message Age Max Age Hello Time Forward Delay Figure.: BPDU fields as the standard IEEE 8.D(998) states And the Root Path Cost is the cumulative cost of all the links leading to the Root Bridge and it is transmitted in the BPDUs. The Root Path Cost is used and stored twice: at the bridge and at the port level. The former simply tells the cost of that bridge to the current Root bridge obviously through the Root port. The latter indicates the Root Path Cost of the Designated bridge for the segment where the Root port is attached to. Therefore, a Designated port stores the value of the local bridge, while Root and Alternate ports have the Designated neighbor cost value. Each physical port has an assigned cost and this depends on the type of link. Generally, the higher the bandwidth of a link, the lower the cost of transporting data across it. The original IEEE 8.D standard defined Path Cost as Mbps divided by the link bandwidth in megabits per second. Modern networks commonly use Gigabit Ethernet and OC-48 ATM, which are both either too close to or greater than the maximum scale of Mbps. The IEEE now uses a nonlinear scale for Path Cost, as shown in table.. Table.: IEEE 8.D link costs Link Speed Recommended value 4Mb/s 5 Mb/s Mb/s 9 Gb/s 4 Gb/s 5

16 The Message Age field carries the approximated age of the information included in the BPDU (this is addressed in chapter 3). When this age is higher than the field Max Age, it is discarded. The last three fields are the values of the protocol operation timers and the Root transmits them on the BPDUs in order to make every node work with the same timer values...3 Protocol operation One of the basis of the protocol in order to agree in a logical tree-shaped topology is the BPDU exchange. A bridge only sends BPDUs in two cases: The bridge is the Root. Then it sends periodical BPDUs in order to keep the network alive and aware of the Root existence. Note that bridges expect periodical BPDUs and when this receptions stop the bridges consider the tree topology to be outdated. Forward a received BPDU. Actually, these are the messages created in the Root, that are forwarded by the rest of the bridges through all the tree. Note that bridges do not exactly forward frames but create new ones based on its current state. However, when an updating BPDU arrives at a port, its information is disseminated in form of new BPDUs (this procedure seems to be forwarding but it is not) This results in flooding the network of BPDUs because the messages created by the real Root are forwarded by all other bridges through all the tree branches. Therefore, all BPDUs that are forwarded contain information about the Root that has originated them. At the network start up all bridges think they are the Root and send their own BPDUs. Only those messages informing about the real Root remain being forwarded. The others are eventually considered inferior to the real Root information, not forwarded any more and hence disappear from the network...3. BPDU reception When a bridge receives a BPDU and considers it as updating, it first saves it into a port database where all last accepted BPDUs are stored. This database is then used when the bridge reconfigures because it chooses the roles of its ports depending on this information. If one of the BPDUs is not accepted by the receiving bridge, it is not stored and then disappears from the network. Therefore, when a bridge receives a BPDU it performs depending on the conveyed information: Superior information. The port database stores inferior information than the just received. The bridges realizes that its state must be updated and executes a reconfiguration procedure. Repeated information. Nothing is done because the received BPDU is repeating the information received in the last BPDU at the same port (the port database already contains the received information). Inferior information. If a port receives a BPDU inferior to the stored in the port database, it is discarded. 6

17 ..3. Reconfiguration procedure The reconfiguration procedure that a bridge executes when an updating BPDU is received depends on the information that ports database store. It is exactly when this database changes (in case of superior BPDU received) that the bridge must be reconfigured. The process includes the following steps: Selection of the Root bridge. Among all the stored Root bridges in the port database, the best (lower value) is selected as the believed Root bridge. It may be the real one or just an intermediate value that will be considered the Root until the definitive information gets the bridge. Selection of the Root port. Once the Root bridge is decided, the path to this Root with the lowest cost is selected as the Root port. Selection of the Designated ports. Among the rest of the ports, those with a inferior stored BPDU are considered Designated ports. Note that when a port is elected as Designated the same bridge information is saved in the port database, hence removing the previous inferior BPDU. Selection of the Alternate ports. The rest of the active ports, with superior port database information, are selected as Alternate ports. All decisions are taken based on the port database information. It may happen that in the Root port selection two ports have the same cost to the root. Then a sequence of comparisons is used in order to untie the circumstance. The winner port must have the:. the lowest Root Bridge ID, or. the lowest Root Path Cost to Root Bridge, or 3. Lowest Sender Bridge ID, or 4. Lowest Sender Port ID The role transition that the reconfiguration procedure carries out is shown in the following tables. The role transitions from the vertical column to the horizontal row. The first one, table., shows the port role transitions that can occur after a BPDU is received in the same port. The easiest case is when the BPDU is repeated and no information is updated, then all roles remain the same. When a superior BPDU is received, it depends on the level of this superiority. If it is the best BPDU of the bridge so far, bridge superior, the port is selected as Root port. If it is just superior locally in that port (it is better than the last received), the decision depends on the previous role. A Designated turns into Alternate because the sender has now better information. An Alternate becomes Designated because the sender has now inferior information. Note that this last situation is an exchange of roles between a Designated and an Alternate belonging to the same segment. Not only BPDUs received at the same port can affect the port role. Table.3 indicates the role transitions in case of a bridge superior BPDU is received in another port. Note that this message is considered to be the best received so far in the bridge, so this may completely reconfigure the port roles. In this case, all ports can become Designated or alternate depending on how superior is the 7

18 received BPDU in the other port. For example, if the message informs about a new Root, the receiving port becomes the new Root port and the rest are now Designated. Another decision that can be resulted from a BPDU reception is that the current bridge becomes the root. In this case all bridges are elected as Designated as table.4 shows. Table.: Roles transition due to BPDU received in the same port Root Designated Alternate Root repeated superior bridge superior inferior Designated bridge superior repeated superior Inferior Alternate bridge superior inferior repeated superior Table.3: Roles transition due to BPDU received in other ports Root Designated Alternate Root bridge superior bridge superior Designated bridge superior bridge superior Alternate bridge superior bridge superior Table.4: Roles transition when the bridge becomes the Root Root Designated Alternate Root Bridge is the Root Designated Bridge is the Root Alternate Bridge is the Root..3.3 Timeout of stored information A part from the reception of updating BPDUs there is another situation that also results in a reconfiguration execution. When a bridge receives a BPDU and stores its information, it also starts a timer that counts the age if the information. This timer is set to a Message Age Timer value (refer to chapter 3 for more information). If the next periodical BPDU is received after a while, the timer is restarted. However, a problem like a failure may happen and the bridge stops receiving BPDUs through that port. Since the timer is not restarted, it expires. This means that the stored information is not valid any more and a reconfiguration procedure is executed. The bridge always selects the ports 8

19 where the timer has expired as Designated ports. Note that only Root ports and Alternate ports receive BPDUs (both from another Designated port), hence only these two kinds of ports can expire information. If an Alternate times out, it becomes Designated. If a Root port times out, this means that the information from the Root is not valid, and the whole bridge reconfigures believing that it is the new Root, and selects all os its ports as Designated Pseudo-code In order to summarize the protocol actions, a simple pseudo-code is provided following. RECEIVED BPDU if repeated forward BPDUs to Designated ports restart Message Age Timer if inferior if received in Designated reply BPDU else --- if superior store information in port database start Message Age Timer reconfigure bridge Root Bridge selection Root Port selection Designated/Alternate Port selection TIMEOUT OF PORT INFORMATION if Root Port new Root Bridge all ports are Designated else port is Designated PERIODICAL BPDU TRANSMISSION if Root bridge send periodical BPDU..4 Initial convergence example At network start up, no active topology is ready and bridges need to start talking each other in order to agree. This first action is called initial convergence and it is the basic example to show how STP works. This section is divided into different steps where each one describes a single election that bridges carry out after a BPDU receipt. They choose a Root bridge, a Root port and some Designated ports every time they receive an updating BPDU. This is exactly the chronological order of the actions after a single BPDU receipt. First, based 9

20 on the comparison between the stored information in the port database and the arriving BPDU, a Root bridge is elected. After, the closest port to this Root bridge is elected as Root port. And finally Designated ports are chosen among the rest. This is exactly the reconfiguration procedure explained in the previous section. Note that the following points explain these processes separately just in order to make the guidance easier. Do not get confused thinking that first all network agrees in a Root bridge by sending BPDUs and then more BPDUs are sent to select the role ports. In the real protocol operation these three processes run at the same time and the decisions are made at every received BPDU. The idea is that bridges, at every updating BPDU, reconfigure themselves depending on the knowledge of the network they have at the given moment (which can be non-definitive). For example, at the beginning the closest bridges to the Root would know about it, and they already choose the correct Root ports. Other bridges also reconfigure because the information they receive may be better than the stored, but it is not the definitive. Once the information from the Root has reached all paths, all nodes know all about the Root, and the tree is configured. It is very important to really understand that these procedures run in parallel and that all advance at each received BPDU. In order to make it more clear, after the description of the three individual processes a global example is presented. This deals with all the steps regarding the three procedures until the tree is fully configured...4. Root Bridge Selection For all bridges in a network to agree on a loop-free topology, a common reference must exist to be used as a guide. This reference point is the Root Bridge and all bridges must eventually know it. This reference point is the bridge with the lowest Bridge ID. In case of equal priorities, the bridge with the lowest MAC address will be elected as Root. But tuning the priority filed, the Root election can be biased in order to make a fixed Root election. When a bridge first powers up, it has a narrow view of its surroundings and assumes that it is the Root Bridge itself (this belief probably changes as other bridges also participate). Then, the election process proceeds as follows. Every bridge begins by sending BPDUs with a Root Bridge ID equal to its own Bridge ID (each one believes to be itself the Root). Received BPDU messages are analyzed to see if a better Root Bridge is being announced. A Root Bridge is considered better if the Root Bridge ID value is lower than the known so far. The bridge then is required to advertise the new Root Bridge ID in its own BPDU messages that sends through the Designated ports to other neighbors. Sooner or later, the election converges and all bridges agree on the same Root Bridge. Before this is achieved, many intermediate decisions may be done, and a bridge that is located maximally far from the new Root receives many BPDUs informing of other Roots before realizing about the real good information. As an example, consider the network shown in Fig..3. The boxes near the bridges represent the Bridge state at that moment. It includes the Bridge ID

21 Bridge ID: Root ID: Bridge ID: Root ID: 6 Bridge ID: Root ID: Bridge ID: Root ID: Bridge ID: Root ID: Bridge ID: Root ID: 6 6 Bridge ID: Root ID: (t ) Bridge ID: Root ID: 4 4 Bridge ID: Root ID: Bridge ID: Root ID: 5 Bridge ID: Root ID: 5 5 Bridge ID: Root ID: Bridge ID: Root ID: 6 Bridge ID: Root ID: 6 3 Bridge ID: Root ID: 4 Bridge ID: Root ID: (t ) 5 3 Bridge ID: Root ID: 5 3 Figure.3: Root Bridge election 3 Bridge ID: Root ID: Bridge ID: Root ID: 3 5 Bridge ID: Root ID: 5 Root ID (t 3) 4 Bridge ID: Root ID: 4

22 (this is fixed) and the believed Root ID (this may converge to the Real Root). Bridge is chosen as Root because it has the lower MAC address (assuming equal priorities). At the startup, each bridge will send BPDUs claiming to be the Root (t ) but only those from bridge will continue being forwarded (t ). The good information about the Root traverses all the network until all nodes know about bridge existence (t 3 ). At that point all bridges agree in bridge as the Root bridge of the network. Note that the information from must traverse the network until all ports of all bridges know about the future Root. For example the tree is not fully configured until 5 receives from 4 a BPDU informing about, and vice-versa. Note that until the good information about has not traversed the complete network, some bridges may assume wrong configurations. This transient situation finishes once the tree converges to the stable configuration. At his point the tree is formed. For example bridge 4 believes that is the Root (t ) after receiving the initial BPDU from it. However, it finally reconfigures accepting bridge as the new Root bridge (t 3 ). In addition, if a new bridge with a lower Bridge Priority powers up, it begins advertising itself as the Root Bridge. Because the new bridge have a lower Bridge ID, all the bridges soon reconsider and record it as the new Root Bridge. This also can happen if the new bridge has a Bridge Priority equal to that of the existing Root Bridge but has a lower MAC address. Now, let us recall the statement of the three processes running in parallel. As the Root bridge election procedure advances, also the other two procedures do so. For example, at (t ) the bridges closest to the Root already know which are their definitive Root port, while the furthest still have transient configurations...4. Root Port Selection With the previous mechanism, a Root bridge is elected as a reference point for the entire bridged network. Each one of the other bridges must figure out where it is in relation to this reference. This action is performed by selecting only one Root Port on each non-root bridge. The Root Port always points toward the current Root Bridge and the selected one is the port with the lowest cost to the bridge. STP uses the concept of cost to determine many things. Selecting a Root Port involves evaluating the Root Path Cost. The Root Path Cost value of each bridge is determined in the following manner:. The Root Bridge sends out a BPDU with a Root Path Cost value of because its ports sit directly on the Root Bridge.. When the next-closest neighbor receives the BPDU, it adds the Path Cost of its own port where the BPDU arrived. (This is done as the BPDU is received.) 3. The neighbor sends out BPDUs with this new cumulative value as the Root Path Cost. 4. The Root Path Cost is incremented by the ingress port Path Cost as the BPDU is received at each bridge down the line.

23 After incrementing the Root Path Cost, a bridge also records the value in its memory. When a BPDU is received on another port and the new Root Path Cost is lower than the previously recorded value, this lower value becomes the new Root Path Cost. In addition, the lower cost tells the bridge that the path to the Root Bridge must be better using this port than it was on other ports. Hence the new Root port is the port containing the lowest Root Path Cost. Bridge ID: Root ID: Bridge ID: Root ID: R R Bridge ID: Root ID: 6 6 R 3 R Bridge ID: Root ID: 3 R 5 Bridge ID: Root ID: Bridge ID: Root ID: 5 Root Path Cost R 4 link cost : R Root Port D Designated Port Bridge ID: Root ID: 4 Figure.4: Root Ports election Fig..4 shows the same example as before, with the added update of the root ports election. The figure shows the definitive Root ports that each bridge choose. Taking as en example bridge 3, it is clear to see the the Root port is the one receiving the lower Root Path Cost value in the port directly connected to bridge. Another case happens in bridge 5, which receives the same cost value of in two different ports. In this case the the selected Root port is decided after applying the 4- steps sequence described above (port connected to bridge 3 is better than port connected to bridge 6). As in the Root bridge election procedure, bridges may take intermediate decisions and select as Root port an interface that is not going to be the final configuration. Only when all BPDUs from the real Root span all the network, the final Root ports are selected Designated Port Selection Once the Root bridge is elected as main reference Root Bridge and all root ports towards it are decided, loops may still exist since no ports have been blocked. In order to break them, STP makes an additional computation to identify only 3

24 Bridge ID: Root ID: Bridge ID: Root ID: D R : D D D D R :6 Bridge ID: Root ID: 6 D 6 R Bridge ID: Root ID: : A R D : D :3 A 3 Bridge ID: Root ID: D 3 R A :5 A 5 :5 Bridge ID: Root ID: 5 Root Path Cost : Bridge ID R 4 :4 D link cost : R Root Port D Designated Port A Alternate Port Bridge ID: Root ID: 4 Figure.5: Designated Ports election one Designated Port for each network segment. Suppose that two or more bridges have a port connected to a common network segment. If a frame appears on that segment, all the bridges attempt to forward it to its destination (this bad behavior is the consequence of a bridging loop). Only the port elected as Designated for that segment should forward traffic. Then, bridges choose a Designated Port based on the lowest cumulative Root Path Cost to the Root Bridge. For example, in a segment with two connected bridges (see bridges and 3 in Fig..5) all of them must agree on who the Designated is. Bridge sends BPDUs with a Root Path Cost of, as well as 3. This is a case of tie, then the decision process steps forward in the 4-step sequence and decides that is the Designated because its Bridge ID is lower. The port that looses is considerd Alternate and turned into Blocking, thus the loop is broken. Note that this decision is individually taken by each bridge. receives a BPDU from 3 and sees that it () has a better priority vector (then it considers itself as the Designated port). Similarly, 3 checks that the information from is better than its own, and accepts as the Designated for that segment. A similar situation happens between bridges -, 4-5 and Global example The previous sections describe the three processes that are run in parallel in the STP operation. The following text now aims at putting everything together in order to give a complete explanation of the simultaneous decisions taken. The objective is to clarify how the protocol really performs when it creates the tree 4

25 Root ID: Cost: Root ID: Cost: Root ID: Cost: :: :: :: :: :: 6:6: 6 Root ID: 6 Cost: Root ID: Cost: 6 Root ID: Cost: :: 6:6: :: 6:: 3:3: :: 3:3: 3 3:3: 3:: 3 3:: :: :: (t ) :: 4:4: 4 4:4: 5:5: 5:5: 5:3: Root ID: Cost: Root ID: Cost: Root ID: Cost: x Root ID: Cost: Root ID: 3 Cost: x 3 Root ID: Cost: Root ID: 4 Cost: 4:: 6 5:5:5 5:: 5 Root ID: 5 Cost: Root ID: Cost: x 5 Root ID: Cost: Bridge ID : Root ID : Cost x Root Port Designated Port Alternate Port Root ID: Cost: :: Root ID: Cost: Root ID: Cost: x :: (t ) :: Root ID: Cost: Root ID: Cost: x 4 3 4:4: Root ID: Cost: Root ID: Cost: 6 5:3: 5 Root ID: 3 Cost: Root ID: Cost: x x 5 Root ID: Cost: Figure.6: STP operation in building the tree 5 (t 3) 4 Root ID: Cost: (t 4) 4 Root ID: Cost:

26 topology. The same example network is used and a set of temporal steps guide the process (each one of these steps is considered as a transmission/reception of a BPDU). Fig..6 shows the mentioned steps. At (t ), and according to the initial configuration of a bridge when it is switched on, everyone believes to be the Root and all send the corresponding BPDUs through its Designated ports (all ports are designated in the Root bridge). These BPDUs contain the same value in the fields Bridge ID and Root ID and equal to own bridge Id. All Root Path Cost are in this stage as the figure shows. Once the BPDUs sent in (t ) are received in the neighbor bridges, some reconfigurations occur (t ). The nodes closest to the future Root (bridge, it has the lower MAC address) receive information directly from. They set the receiving ports as Root ports, because it is the best BPDU they have received so far, and configure all others as Designated ports. Then, all bridges send BPDUs with the new information in order to disseminate that is the Root. Note that bridges 4 and 5 also reconfigure themselves but with still transient configuration. For example, bridge 4 believes that 3 is the Root (after receiving an updating BPDU from 3 telling that 3 is the Root). Now 4 sets its Root port as the port connecting to bridge 3, and all others as Designated. Afterwards, 4 forwards this information through its Designated ports, but it may be discarded if it arrives to nodes that already know about. Bridge 5 behaves similarly. The third step (t 3 ) shows very clear examples of when a port is elected as Designated in a concrete network segment. Let us take the link between and. In (t ) both bridges send BPDUs telling about the new Root. Now these information is received by each other and the Designated port election is executed. Bridge decides that it has the best priority vector and puts its port into Designated because, even though the Root ID and the cost is the same, the Bridge ID is lower. In parallel, bridge also accepts that has better information and selects its port as Alternate (before that, both ports are Designated and this creates a loop that must be broken). The same situation occurs between bridges and 3, and 5 and 6. However, the last two agree in the decision based on non-definitive information because the BPDU that 5 sends tells about 3 as Root. In this example setting the Designated port in 6 and the Alternate port in 5 results in the final configuration because it coincides with the final one. Note the situation in bridges 4 and 5. They receive BPDUs telling about Root and 3 respectively. Since they already know about, they consider these ports as Designated and send new BPDUs (now telling about ). The loop is not yet broken because only one of the ports can be Designated. Finally, in (t 4 ) bridges 4 and 5 already receive the BPDUs about in the last ports and decide to elect as Designated the port of bridge 4, and Alternate the port of bridge 5. Now the tree is finally configured because all information from has already reached all ports of the network. As next section explains, the Designated and Root ports transition to forwarding, and Alternate to blocking. Forming the forwarding tree depicted in the last step of the figure..3 Topology activation The previous section. provides the description of the port roles and how these are decided depending on received messages or information timeouts. In 6

27 the reconfiguration procedure that a bridge executes after receiving an updating BPDU, roles are assigned to ports. Depending on the role and how these transition, the port state of the port is decided..3. Port States The state of the port is selected based on the state diagram shown in Fig..7. This step is the real tree activation of the tree because it is exactly here when the ports are really made operative or blocked. Disabled Forwarding D/R to A D/R to A timer Learning timer by management by protocol operation Blocking D/R to A A to D/R Listening Figure.7: STP port state diagram When a port turns into a state, no immediate actions are executed. These states just indicate which of the following bridge procedures are activated. Forwarding Procedure. If it is activated the port forwards data frames from and to the segment it is connected to. Learning Procedure. If it is activated, the port learns MAC addresses from the incoming frames in the segment it is connected to. Tree Computation Procedure. If it is activated, the port processes the BPDUs that are read from the segment it is connected to. The following table shows which of the previous procedures are active in each one of the states. Regarding the processing of BPDUs, all ports take part in the tree computation except the Disabled (note that this is totally inactive). This means that even a blocked port processes received BPDUs because updating information may arrive. One of the main operations of the bridge is the forwarding of data frames. Only the forwarding state is activated in that sense. Therefore only those ports that are finally transitioned to Forwarding can receive and transmit data frames. Note that only Root and Designated ports must be in this state. The other main action that a bridge executes is the MAC address learning. In this case there are two states that run this procedure: the Forwarding and the Learning state. This last one is introduced in order to avoid a big flooding of unknown unicast frames 7