Chapter 11 - Switching and Virtual LANs.pdf

Full Transcript

Chapter 11 Switching and Virtual LANs THE FOLLOWING COMPTIA NETWORK+ EXAM OBJECTIVES ARE COVERED IN THIS CHAPTER: Domain 2.0 Network Implementation 2.2 Given a scenario, configure switching technologies and features. Virtual Local Area Network (VLAN) VLAN database Switc...

Chapter 11 Switching and Virtual LANs THE FOLLOWING COMPTIA NETWORK+ EXAM OBJECTIVES ARE COVERED IN THIS CHAPTER: Domain 2.0 Network Implementation 2.2 Given a scenario, configure switching technologies and features. Virtual Local Area Network (VLAN) VLAN database Switch Virtual Interface (SVI) Interface configuration Native VLAN Voice VLAN 802.1Q tagging Link aggregation Speed Duplex Spanning tree Maximum transmission unit (MTU) Jumbo frames Domain 5.0 Network Troubleshooting 5.3 Given a scenario, troubleshoot common issues with network services. Switching issues STP Network loops Root bridge selection Port roles Port states 5.4 Given a scenario, troubleshoot common performance issues. Layer 2 switching is the process of using the hardware addresses of devices on a LAN to segment a network. Because you've got the basic ideas down, I'm now going to focus on the more in-depth particulars of layer 2 switching and how it works. You already know that switching breaks up large collision domains into smaller ones and that a collision domain is a network segment with two or more devices sharing the same bandwidth. A hub network is a typical example of this type of technology. But because each port on a switch is actually its own collision domain, you can create a much better Ethernet LAN by simply replacing your hubs with switches! Switches truly have changed the way networks are designed and implemented. If a pure switched design is properly implemented, it will result in a clean, cost-effective, and resilient internetwork. In this chapter, we'll survey and compare how networks were designed before and after switching technologies were introduced. Routing protocols like Routing Information Protocol (RIP), which you learned about in Chapter 10, “Routing Protocols,” employ processes for preventing network loops from occurring at the Network layer. This is all good, but if you have redundant physical links between your switches, routing protocols won't do a thing to stop loops from occurring at the Data Link layer. That's exactly the reason Spanning Tree Protocol was developed —to put a stop to loops taking place within a layer 2 switched network. The essentials of this vital protocol, as well as how it works within a switched network, are some of the important subjects that we'll cover thoroughly in this chapter. To finish up this chapter, you're going to learn exactly what a VLAN is and how VLAN memberships are used in a switched network as well as how trunking is used to send information from all VLANs across a single link. Good stuff! To find Todd Lammle CompTIA videos and practice questions, please see www.lammle.com. Networking Before Layer 2 Switching Because knowing the history of something really helps with understanding why things are the way they are today, I'm going to go back in time a bit and talk about the condition of networks before switches and the part switches have played in the evolution of corporate LANs by helping to segment them. For a visual of how a typical network design looked before LAN switching, check out the network in Figure 11.1. FIGURE 11.1 A network before switching The design in Figure 11.1 was called a collapsed backbone because all the hosts involved had to go to the corporate backbone to reach any network services—both LAN and mainframe. Going back even further, before networks like the one shown in Figure 11.1 had physical segmentation devices such as routers and hubs, there was the mainframe network. This type of network comprised mainframe controllers made by IBM, Honeywell, Sperry, DEC, and so on and dumb terminals that connected into the controller(s). Any remote sites were connected to the mainframe with bridges. And then the PC began its rise to stardom, and the mainframe was connected to an Ethernet or Token Ring LAN where the servers were installed. These servers were usually OS/2 or LAN Manager because this was “pre-NT.” Each floor of a building ran either coax or twisted-pair wiring to the corporate backbone, which was then connected to a router. PCs ran an emulating software program that allowed them to connect to mainframe services, giving those PCs the ability to access services from the mainframe and LAN simultaneously. Eventually, the PC became robust enough to allow application developers to port applications more effectively than they ever could before—an advance that markedly reduced networking prices and enabled businesses to grow at a much faster rate. Moving forward to when Novell rose to popularity in the late 1980s and early 1990s, OS/2 and LAN Manager servers were by and large replaced with NetWare servers. This made the Ethernet network even more popular because that's what Novell 3.x servers used to communicate with client-server software. So basically, that's the story about how the network in Figure 11.1 came into being. But soon a big problem arose with this configuration. As the corporate backbone grew and grew, network services became slower and slower. A big reason for this was that at the same time this huge burst in growth was taking place, LAN services began to require even faster response times. This resulted in networks becoming totally saturated and overwhelmed. Everyone was dumping the dumb terminals used to access mainframe services in favor of those slick new PCs so they could more easily connect to the corporate backbone and network services. And all this was taking place before the Internet's momentous popularity, so everyone in the company needed to access the corporate network's own, internal services. Without the Internet, all network services were internal, meaning that they were exclusive to the company network. As you can imagine, this situation created a screaming need to segment that single, humongous, and now plodding corporate network, which was connected together with sluggish old routers. How was this issue addressed? Well, at first, Cisco responded by simply creating faster routers (no doubt about that), but still more segmentation was needed, especially on the Ethernet LANs. The invention of Fast Ethernet was a very good and helpful thing, yet it too fell short of solving that network segmentation need. But devices called bridges did provide relief, and they were first used in the networking environment to break up collision domains. Sounds good, but only so much—bridges were sorely limited by the number of ports and other network services they could provide, and that's when layer 2 switches came to the rescue. These switches saved the day by breaking up collision domains on each and every port—like a bridge—but switches could provide hundreds of ports! This early, switched LAN looked like the network pictured in Figure 11.2. As you can see here, each hub was placed into a switch port—an innovation that vastly improved the network. So now, instead of each building being crammed into the same collision domain, each hub became its own separate collision domain. Yet still, as is too often the case, there was a catch—switch ports were still very new and, therefore, super expensive. Because switches were so cost prohibitive, simply adding a switch into each floor of the building just wasn't going to happen—at least not yet. But thanks to whomever you choose to thank for these things, the switch price tag has dropped dramatically; now, having every one of your users plugged into a switch port is a really good solution, and cost effective too! So there it is—if you're going to create a network design and implement it, including switching services is a must. A typical, contemporary, and complete switched network design/implementation would look something like Figure 11.3. FIGURE 11.2 The first switched LAN FIGURE 11.3 The typical switched network design “But wait—there's still a router in there!” you say. Yes, it's not a mirage—there is a router in there. But its job has changed quite a bit. Instead of performing physical segmentation, it now creates and handles logical segmentation. Those logical segments are called VLANs, and no worries, I promise to explain them thoroughly throughout the rest of this chapter. Switching Services Bridges use software to create and manage a filter table, but switches use application- specific integrated circuits (ASICs) to accomplish this. Even so, it's still okay to think of a layer 2 switch as a multiport bridge because their basic reason for being is the same: to break up collision domains. Layer 2 switches and bridges are faster than routers because they don't take up time looking at the Network layer header information. Instead, they look at the frame's hardware addresses before deciding to forward, flood, or drop the frame. Switches create private, dedicated collision domains and provide independent bandwidth on each port, unlike hubs. Figure 11.4 shows Six hosts connected to a switch —all running 100 Mbps full duplex to the server. Unlike with a hub, each host has full- duplex, 100 Mbps of dedicated communication to the server. Common switchports today can pass traffic at 10/100/1000 Mbps depending on the connected device. By default, the switchports are set to auto-configure. FIGURE 11.4 Switches create private domains Layer 2 switching provides the following benefits: Hardware-based bridging (ASIC) Wire speed Low latency Low cost What makes layer 2 switching so efficient is that no modification to the data packet takes place. The device reads only the frame encapsulating the packet, which makes the switching process considerably faster and less error prone than routing processes. And if you use layer 2 switching for both workgroup connectivity and network segmentation (breaking up collision domains), you can create a flatter network design with more network segments than you can with traditional routed networks. Plus, layer 2 switching increases bandwidth for each user because, again, each connection (interface) into the switch is its own collision domain. This feature makes it possible for you to connect multiple devices to each interface—very cool. Coming up, we'll dive deeper into the layer 2 switching technology. Limitations of Layer 2 Switching Because people usually toss layer 2 switching into the same category as bridged networks, we also tend to think it has the same hang-ups and issues that bridged networks do. Keep in mind that bridges are good and helpful things if we design the network correctly, keeping our devices' features as well as their limitations in mind. To end up with a solid design that includes bridges, there are two really important things to consider: You absolutely have to break up the collision domains properly. A well-oiled, functional bridged network is one whose users spend 80 percent of their time on the local segment. So, bridged networks break up collision domains, but remember, that network is really still just one big broadcast domain. Neither layer 2 switches nor bridges break up broadcast domains by default—something that not only limits your network's size and growth potential but can also reduce its overall performance! Broadcasts and multicasts, along with the slow convergence time of spanning trees, can give you some major grief as your network grows. These are the big reasons layer 2 switches and bridges just can't completely replace routers (layer 3 devices) in the internetwork. Bridging vs. LAN Switching It's true—layer 2 switches really are pretty much just bridges that give us lots more ports. But the comparison doesn't end there. Here's a list of some significant differences and similarities between bridges and switches that you need to keep in mind: Bridges are software-based, whereas switches are hardware-based because they use ASIC chips to help make filtering decisions. A switch can be viewed as a multiport bridge. There can be only one spanning-tree instance per bridge, whereas switches can have many. (I'm going to tell you all about spanning trees in a bit.) Switches have a higher number of ports than most bridges. Both bridges and switches forward layer 2 broadcasts. Bridges and switches learn MAC addresses by examining the source address of each frame received. Both bridges and switches make forwarding decisions based on layer 2 addresses. Three Switch Functions at Layer 2 There are three distinct functions of layer 2 switching—you need to know these! They are as follows: Address learning Forward/filter decisions Loop avoidance The next three sections cover these functions in detail. Address Learning Layer 2 switches and bridges are capable of address learning; that is, they remember the source hardware address of each frame received on an interface and enter this information into a MAC database known as a forward/filter table. But first things first— when a switch is initially powered on, the MAC forward/filter table is empty, as shown in Figure 11.5. FIGURE 11.5 Empty forward/filter table on a switch When a device transmits and an interface receives a frame, the switch places the frame's source address in the MAC forward/filter table, which allows it to remember the interface on which the sending device is located. The switch then has no choice but to flood the network with this frame out of every port except the source port because it has no idea where the destination device is actually located. If a device answers this flooded frame and sends a frame back, then the switch will take the source address from that frame and place that MAC address in its database as well, thereby associating the newly discovered address with the interface that received the frame. Because the switch now has both of the relevant MAC addresses in its filtering table, the two devices can make a point-to-point connection. The switch doesn't need to flood the frame as it did the first time because now the frames can and will be forwarded only between the two devices recorded in the table. This is exactly the thing that makes layer 2 switches better than hubs, because in a hub network, all frames are forwarded out all ports every time—no matter what. This is because hubs just aren't equipped to collect, store, and draw upon data in a table as a switch is. Figure 11.6 shows the processes involved in building a MAC database. FIGURE 11.6 How switches learn hosts' locations In this figure, you can see four hosts attached to a switch. When the switch is powered on, it has nothing in its MAC address forward/filter table (just as in Figure 11.5). But when the hosts start communicating, the switch places the source hardware address of each frame in the table along with the port that the frame's address corresponds to. Let me give you a step-by-step example of how a forward/filter table becomes populated: 1. Host A sends a frame to Host B. Host A's MAC address is 0000.8c01.000A, and Host B's MAC address is 0000.8c01.000B. 2. The switch receives the frame on the E0/0 interface and places the source address in the MAC address table, associating it with the port it came in on. 3. Because the destination address is not in the MAC database, the frame is forwarded (flooded) out all interfaces—except the source port. 4. Host B receives the frame and responds to Host A. The switch receives this frame on interface E0/1 and places the source hardware address in the MAC database, associating it with the port it came in on. 5. Host A and Host B can now make a point-to-point connection, and only the two devices will receive the frames. Hosts C and D will not see the frames, nor are their MAC addresses found in the database, because they haven't yet sent a frame to the switch. Oh, by the way, it's important to know that if Host A and Host B don't communicate to the switch again within a certain amount of time, the switch will flush their entries from the database to keep it as current as possible. Forward/Filter Decisions When a frame arrives at a switch interface, the destination hardware address is compared to the forward/filter MAC database and the switch makes a forward/filter decision. In other words, if the destination hardware address is known (listed in the database), the frame is sent out only the specified exit interface. The switch will not transmit the frame out any interface except the destination interface. Not transmitting the frame preserves bandwidth on the other network segments and is called frame filtering. But as I mentioned earlier, if the destination hardware address isn't listed in the MAC database, then the frame is flooded out all active interfaces except the interface on which the frame was received. If a device answers the flooded frame, the MAC database is updated with the device's location—its particular interface. So, by default, if a host or server sends a broadcast on the LAN, the switch will flood the frame out all active ports except the source port. Remember, the switch creates smaller collision domains, but it's still one large broadcast domain by default. In Figure 11.7, you can see Host A sending a data frame to Host D. What will the switch do when it receives the frame from Host A? FIGURE 11.7 Forward/filter table If you answered that because Host A's MAC address is not in the forward/filter table, the switch will add the source address and port to the MAC address table and then forward the frame to Host D, you're halfway there. If you also came back with, “If Host D's MAC address was not in the forward/filter table, the switch would have flooded the frame out all ports except for port Fa0/3,” then congratulations—you nailed it! Let's take a look at the output of a show mac address-table command as seen from a Cisco Catalyst switch (the MAC address table works pretty much exactly the same on all brands of switches): Switch#sh mac address-table Vlan Mac Address Type Ports --- ------------- ------ ----- 1 0005.dccb.d74b DYNAMIC Fa0/1 1 000a.f467.9e80 DYNAMIC Fa0/3 1 000a.f467.9e8b DYNAMIC Fa0/4 1 000a.f467.9e8c DYNAMIC Fa0/3 1 0010.7b7f.c2b0 DYNAMIC Fa0/3 1 0030.80dc.460b DYNAMIC Fa0/3 1 0030.9492.a5dd DYNAMIC Fa0/1 1 00d0.58ad.05f4 DYNAMIC Fa0/1 Now suppose the preceding switch received a frame with the following MAC addresses: Source MAC: 0005.dccb.d74b Destination MAC: 000a.f467.9e8c How will the switch handle this frame? The right answer is that the destination MAC address will be found in the MAC address table and the frame will be forwarded out Fa0/3 only. Remember that if the destination MAC address is not found in the forward/filter table, it will forward the frame out all ports of the switch looking for the destination device. Now that you can see the MAC address table and how switches add hosts' addresses to the forward filter table, how do you stop switching loops if you have multiple links between switches? Let's talk about this possible problem in more detail. Loop Avoidance Redundant links between switches can be a wise thing to implement because they help prevent complete network failures in the event that one link stops working. But it seems like there's always a downside—even though redundant links can be extremely helpful, they often cause more problems than they solve. This is because frames can be flooded down all redundant links simultaneously, creating network loops as well as other evils. Here are a few of the problems you can face: If no loop avoidance schemes are put in place, the switches will flood broadcasts endlessly throughout the internetwork. This is sometimes referred to as a broadcast storm. (In real life, it's often referred to in less polite ways that we're not permitted to repeat in print!) Figure 11.8 illustrates how a broadcast can be propagated throughout the network. Pay special attention to how a frame is continually being flooded through the internetwork's physical network media. One way to test the loop avoidance operations of your switch network is to plug one end of a cable into one port and the other end of the same cable into another port. If loop avoidance is not operational, this should cause a big broadcast storm! FIGURE 11.8 Broadcast storm What you see in Figure 11.8 is that a device can receive multiple copies of the same frame because that frame can arrive from different segments at the same time. Figure 11.9 demonstrates how a whole bunch of frames can arrive from multiple segments simultaneously. The server in the figure sends a unicast frame to another device connected to Segment 1. Because it's a unicast frame, Switch A receives and forwards the frame, and Switch B provides the same service—it forwards the unicast. This is bad because it means that the destination device on Segment 1 receives that unicast frame twice, causing additional overhead on the network. You may have thought of this one: The MAC address filter table could be totally confused about the device's location because the switch can receive the frame from more than one link. Worse, the bewildered switch could get so caught up in constantly updating the MAC filter table with source hardware address locations that it might fail to forward a frame! This is called thrashing the MAC table. One of the nastiest things that can happen is having multiple loops propagating throughout a network. This means you end up with loops occurring within other loops, and if a broadcast storm happened at the same time, the network wouldn't be able to perform frame switching at all—it's toast! All of these problems spell disaster (or something like it) and are decidedly ugly situations that just must be avoided or at least fixed somehow. That's where the Spanning Tree Protocol comes into the game. It was developed to solve each and every one of the problems I just told you about. FIGURE 11.9 Multiple frame copies Distributed Switching In a virtual environment such as you might find in many of today's data centers, not only are virtual servers used in the place of physical servers, but virtual switches (software-based) are used to provide connectivity between the virtual systems. These virtual servers reside on physical devices that are called hosts. The virtual switches can be connected to a physical switch to enable access to the virtual servers from the outside world. One of the unique features of these virtual switches is the ability of the switches to span multiple physical hosts. When this is done, the switch is called a distributed switch. This provides connectivity between virtual servers that are located on different hosts, as shown in Figure 11.10. Spanning Tree Protocol Once upon a time, a company called Digital Equipment Corporation (DEC) was purchased and renamed Compaq. But before that happened, DEC created the original version of Spanning Tree Protocol (STP). The IEEE later created its own version of STP called 802.1D. Yet again, it's not all clear skies—by default, most switches run the IEEE 802.1D version of STP, which isn't compatible with the DEC version. The good news is that there is a newer industry standard called 802.1w, which is faster but not enabled by default on any switches. FIGURE 11.10 Distributed switching To begin with, STP's main task is to stop network loops from occurring on your layer 2 network (bridges or switches). It achieves this feat by vigilantly monitoring the network to find all links and making sure that no loops occur by shutting down any redundant ones. STP uses the Spanning Tree Algorithm (STA) to first create a topology database and then search out and destroy redundant links. With STP running, frames will be forwarded only on the premium, STP-picked links. Switches transmit Bridge Protocol Data Units (BPDUs) out all ports so that all links between switches can be found. STP is a layer 2 protocol that is used to maintain a loop-free switched network. STP is necessary in networks such as the one shown in Figure 11.11. In Figure 11.11, you see a switched network with a redundant topology (switching loops). Without some type of layer 2 mechanism to stop network loops, we would fall victim to the problems I discussed previously: broadcast storms and multiple frame copies. Understand that the network in Figure 11.11 would actually sort of work, albeit extremely slowly. This clearly demonstrates the danger of switching loops. And to make matters worse, it can be super hard to find this problem once it starts! FIGURE 11.11 A switched network with switching loops Spanning Tree Port States The ports on a bridge or switch running STP can transition through five different states: Blocking A blocked port won't forward frames; it just listens to BPDUs and will drop all other frames. The purpose of the blocking state is to prevent the use of looped paths. All ports are in a blocking state by default when the switch is powered up. Listening The port listens to BPDUs to make sure no loops occur on the network before passing data frames. A port in listening state prepares to forward data frames without populating the MAC address table. Learning The switch port listens to BPDUs and learns all the paths in the switched network. A port in learning state populates the MAC address table but doesn't forward data frames. Forward delay is the time it takes to transition a port from listening to learning mode. It's set to 15 seconds by default. Forwarding The port sends and receives all data frames on the bridged port. If the port is still a designated or root port at the end of the learning state, it enters the forwarding state. Disabled A port in the disabled state (administratively) does not participate in the frame forwarding or STP. A port in the disabled state is virtually nonoperational. Switches populate the MAC address table in learning and forwarding modes only. Switch ports are usually in either the blocking or forwarding state. A forwarding port is one that has been determined to have the lowest (best) cost to the root bridge. But when and if the network experiences a topology change because of a failed link or when someone adds a new switch into the mix, you'll find the ports on a switch in the listening and learning states. As I mentioned, blocking ports is a strategy for preventing network loops. Once a switch determines the best path to the root bridge, all other redundant ports will be in blocking mode. Blocked ports can still receive BPDUs—they just don't send out any frames. If a switch determines that a blocked port should now be the designated, or root, port, say because of a topology change, the port will respond by going into listening mode and checking all the BPDUs it receives to ensure that it won't create a loop once the port goes back into forwarding mode. STP Convergence Convergence is what happens when all the ports on bridges and switches have transitioned to either forwarding or blocking modes. During this phase, no data will be forwarded until the convergence event is complete. Plus, before data can begin being forwarded again, all devices must be updated. Yes—you read that right: When STP is converging, all host data stops transmitting! So if you want to remain on speaking terms with your network's users (or remain employed for any length of time), you positively must make sure that your switched network is physically designed really well so that STP can converge quickly and painlessly. Figure 11.12 demonstrates a really great way to design and implement your switched network so that STP converges efficiently. FIGURE 11.12 An optimal hierarchical switch design Convergence is truly important because it ensures that all devices are in either the forwarding mode or the blocking mode. But as I've drilled into you, it does cost you some time. It usually takes 50 seconds to go from blocking to forwarding mode, and I don't recommend changing the default STP timers. (You can adjust those timers if you really have to.) By creating your physical switch design in a hierarchical manner, as shown in Figure 11.12, you can make your core switch the STP root. This makes everyone happy because it makes STP convergence happen fast. Because the typical spanning-tree topology's time to convergence from blocking to forwarding on a switch port is 50 seconds, it can create time-out problems on your servers or hosts—like when you reboot them. To address this hitch, you can disable spanning tree on individual ports. Rapid Spanning Tree Protocol 802.1w How would you like to have a good STP configuration running on your switched network (regardless of the brand of switches) but instead of taking 50 seconds to converge, the switched network can converge in about 5 seconds, or maybe even less. How does that sound? Absolutely—yes, we want this! Well then, welcome to the world of Rapid Spanning Tree Protocol (RSTP). RSTP was not designed to be a “brand-new” protocol but more of an evolution of the 802.1D standard, with faster convergence time when a topology change occurs. Backward compatibility was a must when 802.1w was created. The 802.1w is defined in these different port states (compared to 802.1D): Disabled = discarding Blocking = discarding Listening = discarding Learning = learning Forwarding = forwarding To verify the spanning-tree type running on your Cisco switch, use the following command: S1#sh spanning-tree VLAN0001 Spanning tree enabled protocol ieee Root ID Priority 32769 Address 000d.29bd.4b80 Cost 3012 Port 56 (Port-channel1) Hello Time 2 sec Max Age 20 sec Forward Delay 15 sec Bridge ID Priority 49153 (priority 49152 sys-id-ext 1) Address 001b.2b55.7500 Hello Time 2 sec Max Age 20 sec Forward Delay 15 sec Aging Time 15 Uplinkfast enabled Interface Role Sts Cost Prio.Nbr Type --------------------------------------------- Fa0/3 Desg FWD 3100 128.3 Edge Shr Fa0/4 Desg FWD 3019 128.4 Edge P2p Fa0/8 Desg FWD 3019 128.8 P2p Po1 Root FWD 3012 128.56 P2p Since the type output shows Spanning tree enabled protocol ieee, we know we are running the 802.1D protocol. If the output shows RSTP, then you know your switch is running the 802.1w protocol. EXERCISE 11.1 Understanding Switch Functions In this exercise, you will explain the various functions of a switch. 1. Write out the main functions of a switch. 2. Explain how each function operates and the protocols involved. After learning the three functions of a switch from this section, you should be able to articulate how each function operates. Virtual LANs I know I keep telling you this, but I've got to be sure you never forget it, so here I go one last time: By default, switches break up collision domains, and routers break up broadcast domains. Okay, I feel better! Now we can move on. In contrast to the networks of yesterday, which were based on collapsed backbones, today's network design is characterized by a flatter architecture—thanks to switches. So now what? How do we break up broadcast domains in a pure switched internetwork? By creating a virtual local area network (VLAN), that's how! A VLAN is a logical grouping of network users and resources connected to administratively defined ports on a switch. When you create VLANs, you gain the ability to create smaller broadcast domains within a layer 2 switched internetwork by assigning the various ports on the switch to different subnetworks. A VLAN is treated like its own subnet or broadcast domain, meaning that frames broadcasted onto the network are only switched between the ports logically grouped within the same VLAN. So, does this mean we no longer need routers? Maybe yes, maybe no—it really depends on what your specific goals and needs are. By default, hosts in a specific VLAN can't communicate with hosts that are members of another VLAN, so if you want inter-VLAN communication, the answer is yes, you still need a router. VLAN Basics Figure 11.13 shows how layer 2 switched networks are typically designed—as flat networks. With this configuration, every broadcast packet transmitted is seen by every device on the network regardless of whether the device needs to receive that data or not. FIGURE 11.13 Flat network structure By default, routers allow broadcasts to occur only within the originating network, whereas switches forward broadcasts to all segments. Oh, and by the way, the reason it's called a flat network is because it's one broadcast domain, not because the actual design is physically flat. In Figure 11.13, you can see Host A sending out a broadcast and all ports on all switches forwarding it—all except each port that receives it. Now check out Figure 11.14. It pictures a switched network and shows Host A sending a frame with Host D as its destination. What's important to get out of this figure is that the frame is forwarded only out of the port where Host D is located. This is a huge improvement over the old hub networks, unless having one collision domain by default is what you really want. (I'm guessing not!) Okay, you already know that the coolest benefit you gain by having a layer 2 switched network is that it creates an individual collision domain segment for each device plugged into each port on the switch. But as is often the case, new advances bring new challenges with them. One of the biggest is that the greater the number of users and devices, the more broadcasts and packets each switch must handle. And of course, the all-important issue of security and its demands also must be considered—while simultaneously becoming more complicated! VLANs present a security challenge because by default, within the typical layer 2 switched internetwork, all users can see all devices. And you can't stop devices from broadcasting, plus you can't stop users from trying to respond to broadcasts. This means your security options are dismally limited to placing passwords on your servers and other devices. FIGURE 11.14 The benefit of a switched network To understand how a VLAN looks to a switch, it's helpful to begin by first looking at a traditional network. Figure 11.15 shows how a network used to be created using hubs to connect physical LANs to a router. FIGURE 11.15 Physical LANs connected to a router Here you can see that each network is attached with a hub port to the router (each segment also has its own logical network number, even though this isn't obvious looking at the figure). Each host attached to a particular physical network has to match that network's logical network number in order to be able to communicate on the internetwork. Notice that each department has its own LAN, so if we needed to add new users to, let's say, Sales, we would just plug them into the Sales LAN and they would automatically be part of the Sales collision and broadcast domain. This design actually did work well for many years. But there was one major flaw: What happens if the hub for Sales is full and we need to add another user to the Sales LAN? Or, what do we do if there's no more physical space for a new employee where the Sales team is located? Hmmm, well, let's say there just happens to be plenty of room over in the Finance section of the building. That new Sales team member will just have to sit on the same side of the building as the Finance people, and we'll just plug the poor soul into the hub for Finance. Simple, right? So wrong! Doing this obviously makes the new user part of the Finance LAN, which is very bad for many reasons. First and foremost, we now have a major security issue. Because the new Sales employee is a member of the Finance broadcast domain, the newbie can see all the same servers and access all network services that the Finance folks can. Second, for this user to access the Sales network services that they need to get their job done, they would have to go through the router to log into the Sales server—not exactly efficient. Now, let's look at what a switch accomplishes for us. Figure 11.16 demonstrates how switches come to the rescue by removing the physical boundary to solve our problem. It also shows how six VLANs (numbered 2 through 7) are used to create a broadcast domain for each department. Each switch port is then administratively assigned a VLAN membership, depending on the host and which broadcast domain it's placed in. FIGURE 11.16 Switches removing the physical boundary So now if we needed to add another user to the Sales VLAN (VLAN 7), we could just assign the port to VLAN 7 regardless of where the new Sales team member is physically located—nice! This illustrates one of the sweetest advantages to designing your network with VLANs over the old collapsed backbone design. Now, cleanly and simply, each host that needs to be in the Sales VLAN is merely assigned to VLAN 7. Notice that I started assigning VLANs with VLAN number 2. The number is irrelevant, but you might be wondering what happened to VLAN 1. Well, that VLAN is an administrative VLAN, and even though it can be used for a workgroup, Cisco recommends that you use it for administrative purposes only. You can't delete or change the name of VLAN 1, and by default, all ports on a switch are members of VLAN 1 until you actually do change them. Now, because each VLAN is considered a broadcast domain, it's got to also have its own subnet number (refer again to Figure 11.16). And if you're also using IPv6, then each VLAN must also be assigned its own IPv6 network number. So you don't get confused, just keep thinking of VLANs as separate subnets or networks. Let's get back to that “because of switches, we don't need routers anymore” misconception. When looking at Figure 11.16, you can see that there are seven VLANs, or broadcast domains, counting VLAN 1 (not shown in the figure). The hosts within each VLAN can communicate with each other but not with anything in a different VLAN because the hosts in any given VLAN “think” that they're actually in a collapsed backbone, illustrated in Figure 11.15. So what handy little device do you think we need to enable the hosts in Figure 11.16 to communicate with a host or hosts on a different VLAN? You guessed it—a router! Those hosts absolutely need to go through a router, or some other layer 3 device, just as they do when they're configured for internetwork communication (as shown in Figure 11.15). It works the same way it would if we were trying to connect different physical networks. Communication between VLANs must go through a layer 3 device. So don't expect mass router extinction anytime soon! To provide inter-VLAN communication (communication between VLANs), you need to use a router or a layer 3 switch. Quality of Service Before we dive in further into VLANs, I want to make sure that you have a fundamental understanding of QoS and why it is important. Quality of service (QoS) refers to the way the resources are controlled so that the quality of services is maintained. It's basically the ability to provide a different priority for one or more types of traffic over other levels; priority is applied to different applications, data flows, or users so that they can be guaranteed a certain performance level. QoS methods focus on one of five problems that can affect data as it traverses network cable: Delay Dropped packets Error Jitter Out-of-order delivery QoS can ensure that applications with a required bit rate receive the necessary bandwidth to work properly. Clearly, on networks with excess bandwidth, this is not a factor, but the more limited your bandwidth is, the more important a concept like this becomes. QoS allows administrators to predict, monitor, and control bandwidth use to ensure it is available to programs and apps that need it. VLAN Memberships Most of the time, VLANs are created by a system administrator who proceeds to assign switch ports to each one. VLANs of this type are known as static VLANs. If you don't mind doing a little more work when you begin this process, assign all the host devices' hardware addresses into a database so your switches can be configured to assign VLANs dynamically anytime you plug a host into a switch. I hate saying things like “obviously,” but obviously, this type of VLAN is known as a dynamic VLAN. I'll be covering both static and dynamic VLANs next. Static VLANs Creating static VLANs is the most common way to create a VLAN, and one of the reasons for that is because static VLANs are the most secure. This security stems from the fact that any switch port you've assigned a VLAN association to will always maintain it unless you change the port assignment manually. Static VLAN configuration is pretty easy to set up and supervise, and it works really well in a networking environment where any user movement within the network needs to be controlled. It can be helpful to use network management software to configure the ports, but you don't have to use it if you don't want to. In Figure 11.16, each switch port was configured manually with a VLAN membership based on which VLAN the host needed to be a member of—remember, the device's actual physical location doesn't matter one bit as long as the VLAN assignments are correctly configured. Which broadcast domain your hosts become members of is purely up to you. And again, remember that each host also has to have the correct IP address information. For instance, you must configure each host in VLAN 2 into the 172.16.20.0/24 network for it to become a member of that VLAN. It's also a good idea to keep in mind that if you plug a host into a switch, you have to verify the VLAN membership of that port. If the membership is different than what's needed for that host, the host won't be able to gain access to the network services that it needs, such as a workgroup server. Static access ports are either manually assigned to a VLAN or assigned through a RADIUS server for use with IEEE 802.1X. It's easy to set an incorrect VLAN assignment on a port, so using a RADIUS server can help in your configurations. Dynamic VLANs On the other hand, a dynamic VLAN determines a host's VLAN assignment automatically. Using intelligent management software, you can base VLAN assignments on hardware (MAC) addresses, protocols, or even applications that work to create dynamic VLANs. For example, let's say MAC addresses have been entered into a centralized VLAN management application and you hook up a new host. If you attach it to an unassigned switch port, the VLAN management database can look up the hardware address and both assign and configure the switch port into the correct VLAN. Needless to say, this makes management and configuration much easier because if a user moves, the switch will simply assign them to the correct VLAN automatically. But here again, there's a catch—initially, you've got to do a lot more work setting up the database. It can be very worthwhile, though! And here's some more good news: You can use the VLAN Management Policy Server (VMPS) service to set up a database of MAC addresses to be used for the dynamic addressing of your VLANs. The VMPS database automatically maps MAC addresses to VLANs. Identifying VLANs Know that switch ports are layer 2–only interfaces that are associated with a physical port. A switch port can belong to only one VLAN if it is an access port or all VLANs if it is a trunk port, as I'll explain in a minute. You can manually configure a port as an access or trunk port, or you can let the Dynamic Trunking Protocol (DTP) operate on a per-port basis to set the switch port mode. DTP does this by negotiating with the port on the other end of the link. Switches are definitely pretty busy devices. As frames are switched throughout the network, they've got to be able to keep track of all the different port types plus understand what to do with them depending on the hardware address. And remember— frames are handled differently according to the type of link they're traversing. There are two different types of links in a switched environment: access ports and trunk ports. Access Ports An access port belongs to and carries the traffic of only one VLAN. Anything arriving on an access port is simply assumed to belong to the VLAN assigned to the port. Any device attached to an access link is unaware of a VLAN membership—the device just assumes it's part of the same broadcast domain, but it doesn't have the big picture, so it doesn't understand the physical network topology at all. Another good thing to know is that switches remove any VLAN information from the frame before it's forwarded out to an access-link device. Remember that access-link devices can't communicate with devices outside their VLAN unless the packet is routed. And you can only create a switch port to be either an access port or a trunk port—not both. So you've got to choose one or the other, and know that if you make it an access port, that port can be assigned to one VLAN only. Voice Access Ports Not to confuse you, but all that I just said about the fact that an access port can be assigned to only one VLAN is really only sort of true. Nowadays, most switches will allow you to add a second VLAN to an access port on a switch port for your voice traffic; it's called the voice VLAN. The voice VLAN used to be called the auxiliary VLAN, which allowed it to be overlaid on top of the data VLAN, enabling both types of traffic through the same port. Even though this is technically considered to be a different type of link, it's still just an access port that can be configured for both data and voice VLANs. This allows you to connect both a phone and a PC device to one switch port but still have each device in a separate VLAN. If you are configuring voice VLANs, you'll want to configure QoS on the switch ports to provide a higher precedence to voice traffic over data traffic to improve sound quality. Suppose you plug a host into a switch port and users are unable to access any server resources. The two typical reasons this happens is because the port is configured in the wrong VLAN membership or STP has shut down the port because STP thought there was possibly a loop. Trunk Ports Believe it or not, the term trunk port was inspired by the telephone system trunks that carry multiple telephone conversations at a time. So it follows that trunk ports can similarly carry multiple VLANs at a time. A trunk link is a 100 Mbps or 1000 Mbps point-to-point link between two switches, between a switch and router, or even between a switch and server, and it carries the traffic of multiple VLANs—from 1 to 4,094 VLANs at a time. Trunking can be a real advantage because with it, you get to make a single port part of a whole bunch of different VLANs at the same time. This is a great feature because you can actually set ports up to have a server in two separate broadcast domains simultaneously so your users won't have to cross a layer 3 device (router) to log in and access it. Another benefit of trunking comes into play when you're connecting switches. Information from multiple VLANs can be carried across trunk links, but by default, if the links between your switches aren't trunked, only information from the configured VLAN will be switched across that link. Check out Figure 11.17. It shows how the different links are used in a switched network. All hosts connected to the switches can communicate to all ports in their VLAN because of the trunk link between them. Remember, if we used an access link between the switches, this would allow only one VLAN to communicate between switches. As you can see, these hosts are using access links to connect to the switch, so they're communicating in one VLAN only. That means that without a router, no host can communicate outside its own VLAN, but the hosts can send data over trunked links to hosts on another switch configured in their same VLAN. FIGURE 11.17 Access and trunk links in a switched network It's finally time to tell you about the VLAN identification methods. VLAN Identification Methods VLAN identification is what switches use to keep track of all those frames as they're traversing a switch fabric. All of our hosts connect together via a switch fabric in our switched network topology. It's how switches identify which frames belong to which VLANs, and there's more than one trunking method: ISL and 802.1Q. Inter-Switch Link (ISL) Inter-Switch Link (ISL) is a way of explicitly tagging VLAN information onto an Ethernet frame. This tagging information allows VLANs to be multiplexed over a trunk link through an external encapsulation method (ISL), which allows the switch to identify the VLAN membership of a frame over the trunked link. By running ISL, you can interconnect multiple switches and still maintain VLAN information as traffic travels between switches on trunk links. ISL functions at layer 2 by encapsulating a data frame with a new header and cyclic redundancy check (CRC). Of note is that this is proprietary to Cisco switches, and it's used for Fast Ethernet and Gigabit Ethernet links only. ISL routing is pretty versatile and can be used on a switch port, on router interfaces, and on server interface cards to trunk a server. Port Tagging/IEEE 802.1Q Created by the IEEE as a standard method of frame tagging, IEEE 802.1Q works by inserting a field into the frame to identify the VLAN. This is one of the aspects of 802.1Q that makes it your only option if you want to trunk between a Cisco switched link and another brand of switch. In a mixed environment, you've just got to use 802.1Q for the trunk to work! Unlike ISL, which encapsulates the frame with control information, 802.1Q inserts an 802.1Q field along with tag control information, as shown in Figure 11.18. FIGURE 11.18 IEEE 802.1Q encapsulation with and without the 802.1Q tag To meet the exam objectives, it's really the 12-bit VLAN ID that matters, so keep your focus on it. This field identifies the VLAN and can be 2∧12 minus 2 for the 0 and 4,095 reserved VLANs, which means an 802.1Q tagged frame can carry information for 4,094 VLANs. It works like this: You first designate each port that's going to be a trunk with 802.1Q encapsulation. The other ports must be assigned a specific VLAN ID in order for them to communicate. VLAN 1 is the default native VLAN, and when 802.1Q is used, all traffic for a native VLAN is untagged. The ports that populate the same trunk create a group with this native VLAN, and each port gets tagged with an identification number reflecting that membership. Again, the default is VLAN 1. The native VLAN allows the trunks to accept information that was received without any VLAN identification or frame tag. The basic purpose of ISL and 802.1Q frame-tagging methods is to provide inter-switch VLAN communication. Remember that any ISL or 802.1Q frame tagging is removed if a frame is forwarded out an access link—tagging is used internally and across trunk links only! Routing Between VLANs Hosts in a VLAN live in their own broadcast domain and can communicate freely. VLANs create network partitioning and traffic separation at layer 2 of the OSI. As I said when I told you why we still need routers, if you want hosts or any other IP-addressable device to communicate between VLANs, you must have a layer 3 device to provide routing. There are three different approaches to support VLAN routing: You can use a router that has an interface for each VLAN, a router that supports ISL or 802.1Q routing, or a layer 3 enabled switch. What we see in Figure 11.19 is that each router interface is plugged into an access link. This means that each of the routers' interface IP addresses would then become the default gateway address for each host in each respective VLAN. This solution is not scalable, because you will run out of physical interfaces before running out of VLANs. FIGURE 11.19 Router connecting three VLANs together for inter-VLAN communication, with one router interface for each VLAN Instead of using a router interface for each VLAN, you can use one GigabitEthernet interface and run ISL or 802.1q trunking. Figure 11.20 shows how a GigabitEthernet interface on a router will look when configured with ISL or 802.1q trunking. This allows all VLANs to communicate through one interface. This is typically referred to as a router on a stick (ROAS) for Inter-VLAN routing. Here we see one physical interface divided into multiple subinterfaces, with one subnet assigned per VLAN, each subinterface being the default gateway address for each VLAN/subnet. An encapsulation identifier must be assigned to each subinterface to define the VLAN ID of that subinterface. FIGURE 11.20 Router on a stick: single router interface connecting all three VLANs together for inter-VLAN communication I really want to point out that this creates a potential bottleneck, as well as a single point of failure, so your host/VLAN count is limited. To how many? Well, that depends on your traffic level. Keep in mind every packet you route to the router will need to come back through the same link to get to its respective destination. To really make things right, you'd be better off using a higher-end switch and routing on the backplane. But if you just happen to have a router sitting around, configuring this method will get you by. But wait, there's still one more way to go about routing! Instead of using an external router interface for each VLAN, or an external router on a stick, we can configure logical interfaces on the backplane of the layer 3 switch; this is called inter-VLAN routing (IVR), and it's configured with a switched virtual interface (SVI). Figure 11.21 shows how hosts see these virtual interfaces. FIGURE 11.21 With IVR, routing runs on the backplane of the switch, and it appears to the hosts that a router is present. In Figure 11.21, it appears there's a router present, but there is no physical router present as there was when we used router on a stick. The IVR process takes little effort, but it's easy to implement, which makes it very cool! Plus, it's a lot more efficient for inter-VLAN routing than an external router is. Don't worry about learning syntax for configuring routers and switches for the CompTIA Network+ exam. The fundamental design of per interface VLAN, ROAS, and IVR configured with SVI for VLAN routing is the key takeaway. VLAN Trunking Protocol The basic goals of VLAN Trunking Protocol (VTP) are to manage all configured VLANs across a switched internetwork and to maintain consistency throughout that network. VTP allows you to add, delete, and rename VLANs—and information about those actions is then propagated to all other switches in the VTP domain. Here's a list of some of the cool features VTP has to offer: Consistent VLAN configuration across all switches in the network Accurate tracking and monitoring of VLANs Dynamic reporting of added VLANs to all switches in the VTP domain Adding VLANs using plug-and-play Very nice, but before you can get VTP to manage your VLANs across the network, you have to create a VTP server (really, you don't need to even do that since all switches default to VTP server mode, but just make sure you have a server). All servers that need to share VLAN information must use the same domain name, and a switch can be in only one domain at a time. So basically, this means that a switch can share VTP domain information with other switches only if they're configured into the same VTP domain. You can use a VTP domain if you have more than one switch connected in a network, but if you've got all your switches in only one VLAN, you just don't need to use VTP. Do keep in mind that VTP information is sent between switches only via a trunk port. Switches advertise VTP management domain information as well as a configuration revision number and all known VLANs with any specific parameters. But there's also something called VTP transparent mode. In it, you can configure switches to forward VTP information through trunk ports but not to accept information updates or update their VTP databases. If you've got sneaky users adding switches to your VTP domain behind your back, you can include passwords, but don't forget—every switch must be set up with the same password. And as you can imagine, this little snag can be a real hassle administratively! Switches detect any added VLANs within a VTP advertisement and then prepare to send information on their trunk ports with the newly defined VLAN in tow. Updates are sent out as revision numbers that consist of summary advertisements. Anytime a switch sees a higher revision number, it knows the information it's getting is more current, so it will overwrite the existing VLAN database with the latest information. You should know these requirements for VTP to communicate VLAN information between switches: The VTP management domain name of both switches must be the same. One of the switches has to be configured as a VTP server. If a VTP is used, a password must be set. No router is necessary. Now that you've got that down, we're going to delve deeper into the world of VTP with VTP modes. VTP Modes of Operation Figure 11.22 shows you all three different modes of operation within a VTP domain. Server This is the default mode for all Catalyst switches. You need at least one server in your VTP domain to propagate VLAN information throughout that domain. Also important is that the switch must be in server mode for you to be able to create, add, and delete VLANs in a VTP domain. VLAN information has to be changed in server mode, and any change made to VLANs on a switch in server mode will be advertised to the entire VTP domain. In VTP server mode, VLAN configurations are saved in NVRAM on the switch. Client In client mode, switches receive information from VTP servers, but they also receive and forward updates, so in this way they behave like VTP servers. The difference is that they can't create, change, or delete VLANs. Plus, none of the ports on a client switch can be added to a new VLAN before the VTP server notifies the client switch of the new VLAN and the VLAN exists in the client's VLAN database. Also good to know is that VLAN information sent from a VTP server isn't stored in NVRAM, which is important because it means that if the switch is reset or reloaded, the VLAN information will be deleted. Here's a hint: If you want a switch to become a server, first make it a client so it receives all the correct VLAN information, and then change it to a server—so much easier! FIGURE 11.22 VTP modes Transparent Switches in transparent mode don't participate in the VTP domain or share its VLAN database, but they'll still forward VTP advertisements through any configured trunk links. An admin on a transparent switch can create, modify, and delete VLANs because they keep their own database—one they keep secret from the other switches. Despite being kept in NVRAM memory, the VLAN database in transparent mode is actually only locally significant. The whole purpose of transparent mode is to allow remote switches to receive the VLAN database from a VTP-server-configured switch through a switch that is not participating in the same VLAN assignments. Do We Really Need to Put an IP Address on a Switch? The answer is absolutely not! Switches have all ports enabled and ready to rock. Take the switch out of the box, plug it in, and the switch starts learning MAC addresses in the CAM. But since the switches are providing layer 2 services, why do we need an IP address? Because you still need an IP address for in-band management, which is used with your virtual terminals, that's why. Telnet, SSH, SNMP, and so on all require IP addresses to communicate with the switch, in-band, through the network. And remember, since all ports are enabled by default, you need to shut down unused ports or assign them to an unused VLAN. Configuring a switch out-of-band means you're not going through the network to configure the device; you're actually using a port, such as a console port, to configure the switch instead. Most of the time, you'll use the console port upon starting up the switch. After that, all the management will be completed in- band. So now you know that the switch needs a management IP address for in-band management purposes, but exactly where do you want to place it? Conveniently, there's something predictably called the management VLAN interface, and that's clearly your target. It's a routed interface found on every switch, and it's referred to as interface VLAN 1. Good to know that this management interface can be changed, and all manufacturers recommend changing it to a different management interface for security purposes. Yes, you can buy switches that are unmanaged, but you would never ever want to do that for an enterprise network! The only environment in which doing that would make sense is in a home network, but that's about it. Anything you get for an office or larger network absolutely must be a managed switch! With all that in mind, let's get down to configuring a switch now. We'll begin our configuration by connecting into the switch via the console and setting the administrative functions. At this point, we'll also assign an IP address to each switch, but as I said, doing that isn't really necessary to make our network function. The only reason we're going to do that is so we can manage/administer it remotely—in-band—via a protocol like Telnet. Let's use a simple IP scheme like 192.168.10.16/28. And by the way, this mask should be familiar to you. Let's check out the following output: Switch>enable Switch#config t Switch(config)#hostname S1 S1(config)#enable secret todd S1(config)#int f0/15 S1(config-if)#description 1st connection to S3 S1(config-if)#int f0/16 S1(config-if)#description 2nd connection to S3 S1(config-if)#speed 1000 S1(config-if)#duplex full S1(config-if)#line console 0 S1(config-line)#password console S1(config-line)#login S1(config-line)#line vty 0 15 S1(config-line)#password telnet S1(config-line)#login S1(config-line)#int vlan 1 S1(config-if)#ip address 192.168.10.17 255.255.255.240 S1(config-if)#no shut S1(config-if)#exit S1(config)#ip default-gateway 192.168.10.30 S1(config)#banner motd #this is my S1 switch# S1(config)#exit S1#copy run start Destination filename [startup-config]? [enter] Building configuration… [OK] S1# In this output, the first thing to notice is that there aren't any IP addresses configured on the switch's physical interfaces. Since all ports on a switch are enabled by default, there's not really a whole lot to configure. But look again—I configured the speed and duplex of the switch to gigabit, full-on port 16. Most of the time you would just leave these as auto- detect, and I actually recommend doing that. This is not the same technology used between switches to determine the cable type, which is auto Medium Dependent Interface Crossover (MDI-X). My next step is to set the console password for out-of-band management and then the VTY (Telnet) password for in-band management. The next task is to set the default gateway of the switch and banner. So you don't get confused, I want to clarify that the default gateway is used to send management (in-band) traffic to a remote network so you can manage the switch remotely. Understand this is not the default gateway for the hosts—the default gateway would be the router interface address assigned to each VLAN. The IP address is configured under a logical interface, called a management domain or VLAN. You can use default VLAN 1 to manage a switched network just as we're doing here, or you can be smart and opt to use a different VLAN for management. The preceding configuration demonstrates how to configure the switch for local management, meaning that the passwords to log into the switch are right there in the switch's configuration. You can also configure switches and routers to store their usernames and passwords remotely for ease of configuration using an AAA server. Doing this allows you to change the passwords at one device without having to telnet into each device separately to change passwords. To get this done, use the following command: S1(config)#aaa authentication login default This tells the switch to use AAA when Telnet or SSH is used for in-band management. This next command tells the switch to use the AAA server if someone is trying to access the console of the switch: S1(config)#aaa authentication login console So remember, no IP addresses on physical switch interfaces, no routing protocols there either, and so on. We're performing layer 2 switching at this point, not routing! Switch Port Protection There are many features that are available to mitigate switch attacks. In the following sections, we'll examine some of these protections. Port Security Clearly, it's a bad idea to have your switches available for anyone to just plug into and play around with. Security is a big deal—even more of a concern regarding wireless security, so why wouldn't we demand switch security as much, if not more? But just how do we actually prevent someone from simply plugging a host into one of our switch ports—or worse, adding a hub, switch, or access point into the Ethernet jack in their office? By default, MAC addresses dynamically appear in your MAC forward/filter database, but the good news is that you can stop bad guys in their tracks by using port security! Figure 11.23 pictures two hosts connected to the single switch port Fa0/3 via either a hub or access point (AP). FIGURE 11.23 Port security on a switch port restricts port access by MAC address. Port Fa0/3 is configured to observe and allow only certain MAC addresses to associate with the specific port, so in this example, Host A is denied access, but Host B is allowed to associate with the port. By using port security, you can limit the number of MAC addresses that can be assigned dynamically to a port, set static MAC addresses, and—here's my favorite part—set penalties for users who abuse your policy! Personally, I like to have the port shut down when the security policy is violated. Making abusers bring me a memo from their boss explaining why they violated the security policy brings with it a certain poetic justice, which is nice. And I'll also require something like that before I'll enable their port again. Things like this really seem to help people remember to behave! DHCP Snooping A rogue DHCP server (one not under your control that is giving out incompatible IP addresses) can be an annoyance that causes users to be unable to connect to network resources, or it may play a part in several types of attacks. In either case, DHCP snooping is a switch feature that can help to prevent your devices from communicating with illegitimate DHCP servers. When enabled, DHCP snooping allows responses to client requests from only DHCP servers located on trusted switch ports (which you define). When only ports where company DHCP servers are located are configured to be trusted, rogue DHCP servers will be unable to respond to client requests. The protection doesn't stop there, however. The switch will also, over time, develop an IP address–to–MAC address table called the bindings table, derived from “snooping” on DHCP traffic to and from the legitimate DHCP server. The bindings table will alert the switch to any packets that have mappings that do not match the table. These frames will be dropped. The bindings table is also used with ARP inspection, which makes the configuration of DHCP snooping a prerequisite of ARP inspection. ARP Inspection Many on-path (previously known as man-in-the-middle) attacks are made possible by the attacker polluting the ARP cache of the two victims such that their cache maps each other's IP addresses to the MAC address of the attacker, thus allowing the attacker to receive all traffic in the conversation. Dynamic ARP inspection (DAI) is a feature that, when configured, uses the DHCP snooping database of IP address–to–MAC address mappings to verify the MAC address mappings of each frame going through the switch. In this way, any frames with incorrect or altered mappings are dropped by the switch, thus breaking any attacks depending on these bogus mappings. Because it uses the DHCP snooping database, the configuration of DHCP snooping is a prerequisite to enabling DAI. Flood Guard Switches can undergo an attack where some malicious individual floods the switch with unknown MAC addresses. Since switches record all MAC addresses of received frames, the switch will continue to update its MAC table with these MAC addresses until it pushes all legitimate MAC addresses out of the limited space provided for the MAC table in memory. Once this occurs, all traffic received by the switch will be unknown to the switch and it will flood this traffic out of all ports, basically turning the switch into a hub. Now the attacker can connect a sniffer to his port and receive all traffic rather than only the traffic destined for that port as would normally be the case. This attack is shown in Figure 11.24. Flood guard is a feature that can be implemented to prevent this attack. It uses two mechanisms to accomplish this: unknown unicast flood blocking (UUFB) and unknown unicast flood rate-limiting (UUFRL). FIGURE 11.24 Flood guard process The UUFB feature blocks unknown unicast and multicast traffic flooding at a specific port, only permitting egress traffic with MAC addresses that are known to exist on the port. The UUFRL feature applies a rate limit globally to unknown unicast traffic on all VLANs. When these two features are combined, flooding attacks can be prevented in switches that support the features. BPDU Guard When a switch that is unknown to you and not under your control is connected to one of your switches, it can play havoc with your STP topology and may even allow the rogue switch to become the root bridge! As you know, when a switch starts receiving STP BPDUs from a new switch, the information in the BPDU (specifically the switch priority) is used to determine if the switch might be a new root bridge (causing a new election) or if the STP topology should be changed. To prevent this from occurring, a feature called BPDU Guard can be implemented. This feature should be enabled on all switch ports that do not connect to known switches. Since most connections between switches and from the switch to a router are trunk ports, then it is typically enabled on all access ports or ports that connect to end devices. The effect of enabling this feature is simple but effective. The ports on which it is enabled will be shut down when a BPDU is received. While reenabling the port can be done manually, you can also configure the port to wait a period and then reenable itself automatically as well. Root Guard Another feature that can be used to maintain the desired STP topology is called Root Guard. This feature is like BPDU Guard in that it also prevents a new switch from altering the topology. It is applied to all interfaces on the current root bridge and prevents these ports from becoming root ports. Despite the name, root ports are present only on non-root switches and represent the best path back to the root bridge. The feature prevents this by disabling a port if a BPDU is received that, because of its superior priority number, would cause a new root bridge election. It differs from BPDU Guard in that BPDU Guard disables a port where it is configured when any BPDU is received. This would be undesirable behavior on the root bridge as it needs to receive those BPDUs to maintain the topology. So, in summary, it helps to maintain the current root bridge's role as the root bridge. Port Bonding Know that almost all Ethernet networks today will typically have multiple links between switches because this kind of design provides redundancy and resiliency. On a physical design that includes multiple links between switches, STP will do its job and put a port or ports into blocking mode. In addition to that, routing protocols like OSPF and EIGRP could see all these redundant links as individual ones, depending on the configuration, which can mean an increase in routing overhead. We can gain the benefits from multiple links between switches by using port channeling. EtherChannel is a port channel technology that was originally developed by Cisco as a switch-to-switch technique for grouping several Fast Ethernet or Gigabit Ethernet ports into one logical channel. Also important to note is that once your port channel is up and working, layer 2 STP and layer 3 routing protocols will treat those bundled links as a single one, which would stop STP from performing blocking. An additional nice result is that because the routing protocols now see this as only a single link, a single adjacency across the link can be formed—elegant! Figure 11.25 shows how a network would look if we had four connections between switches, before and after configuring port channels. FIGURE 11.25 Before and after port channels Now as usual there's the Cisco version and the IEEE version of port channel negotiation protocols to choose from, and you can take your pick. Cisco's version is called Port Aggregation Protocol (PAgP), and the IEEE 802.3ad standard is called Link Aggregation Control Protocol (LACP). Both versions work equally well, but the way you configure each is slightly different. Keep in mind that both PAgP and LACP are negotiation protocols and that EtherChannel can actually be statically configured without PAgP or LACP. Still, it's better to use one of these protocols to help with compatibility issues as well as to manage link additions and failures between two switches. Cisco EtherChannel allows us to bundle up to eight active ports between switches. The links must have the same speed, duplex setting, and VLAN configuration. In other words, you can't mix interface types and configurations into the same bundle. Here are a few things to remember: Port Channeling/Bonding This refers to combining two to eight Fast Ethernet or Gigabit Ethernet ports together between two switches into one aggregated logical link to achieve more bandwidth and resiliency. EtherChannel This is Cisco's proprietary term for port channeling. PAgP This is a Cisco proprietary port channel negotiation protocol that aids in the automatic creation of EtherChannel links. All links in the bundle must match the same parameters (speed, duplex, VLAN info), and when PAgP identifies matched links, it groups the links into an EtherChannel. This is then added to STP as a single bridge port. At this point, PAgP's job is to send packets every 30 seconds to manage the link for consistency, any link additions and modifications, and failures. LACP (802.3ad) This has the same purpose as PAgP, but it's nonproprietary, so it can work between multivendor networks. Device Hardening A discussion of switch security would be incomplete without discussing device hardening. Actually, this section is applicable to not only the routers and switches in your network but to all devices, including endpoints such as laptops, mobile devices, and desktops. One of the ongoing goals of operations security is to ensure that all systems have been hardened to the extent that is possible and still provide functionality. The hardening can be accomplished both on a physical and logical basis. From a logical perspective, you can do the following: Remove unnecessary applications. Disable unnecessary services. Block unrequired ports. Tightly control the connection of external storage devices and media, if it's allowed at all. Keep firmware updated. Advanced Features of Switches Switches really expand our flexibility when we're designing our networks. The features that we need to cover for the CompTIA Network+ objectives are as follows: Power over Ethernet (PoE)/Power over Ethernet Plus (PoE+) Port mirroring/spanning (local versus remote) Jumbo frames Power over Ethernet (802.3af, 802.3at) Power over Ethernet (PoE and PoE+) technology describes a system for transmitting electrical power, along with data, to remote devices over standard twisted-pair cable in an Ethernet network. This technology is useful for powering IP telephones (Voice over IP, or VoIP), wireless LAN access points, network cameras, remote network switches, embedded computers, and other appliances—situations where it would be inconvenient, expensive, and possibly not even feasible to supply power separately. One reason for this is that the main wiring usually must be done by qualified and/or licensed electricians for legal and/or insurance mandates. The IEEE has created a standard for PoE called 802.3af, and for PoE+ it's referred to as 802.3at. This standard describes precisely how a powered device is detected and also defines two methods of delivering Power over Ethernet to a given powered device. Keep in mind that the PoE+ standard, 802.3at, delivers more power than 802.3af, which is compatible with Gigabit Ethernet with four-wire pairs at 25.5 watts. This process happens one of two ways: either by receiving the power from an Ethernet port on a switch (or other capable device) or via a power injector. You can't use both approaches to get the job done; this can lead to serious trouble, so be sure before connecting! Real World Scenario PoE It would be rare for me not to design a network around PoE. Most of my consulting work is wireless networking, including large outdoor wireless networks. When I design the network, I order equipment based on the amount of power needed to run it, knowing I'll have only a few electrical outlets, or even no outlets if all my equipment is outside. This means that all my switches must run PoE to my access points and wireless bridges and must do this for long distances. For me to accomplish this, I need to order the more expensive, large-scale enterprise switches. If you have devices that need PoE but do not have long- distance connections, you can use lower-end switches, but you must verify that they provide the right amount of power. There was a customer who called me because their network access points were going up and down. The bottom line is that they had purchased less-expensive switches, and there was not enough power to run the equipment. They ended up buying all new switches. So, before you buy a PoE switch, verify that the switch provides the right power for your environment. Figure 11.26 shows an example of a switch that requires PoE injectors to provide PoE to devices versus a switch that provides PoE to any PoE-capable device. A switch that provides PoE makes for a much simpler installation and maintenance in the future! FIGURE 11.26 Non-PoE switch versus a PoE switch As I just said, if you don't have a switch with PoE, then you can use a power injector. Figure 11.27 shows a typical power injector physically installed in a network. Use caution when using an external power injector! Take the time to make sure the power injector provides the voltage level for which your device was manufactured. Because most higher-end switches provide PoE, we don't need to worry about injectors, but if you are adding a wireless bridge into an existing network that has switches without PoE, you need to add a power injector. Figure 11.28 shows a power injector used for a wireless bridge. FIGURE 11.27 An external power injector used for PoE FIGURE 11.28 Wireless bridge power injector Now, let's discuss how we would troubleshoot a network that has a switch in the LAN instead of a hub. Port Mirroring/Spanning (SPAN/RSPAN) Port mirroring, also called Switch Port Analyzer (SPAN) and Remote SPAN, allows you to sniff traffic on a network when using a switch. In Figure 11.29, you can see how a typical switch will read the forward/filter table and send traffic only out the destination port (this is the whole idea of using a switch, so this is good!). FIGURE 11.29 Switches send frames out the destination port only. All good, but a problem with this arises when you need to sniff traffic on the network. Figure 11.29 illustrates this issue; the sniffer isn't seeing data coming from Host A to Host B. To solve this little snag, you can temporarily place a hub between Host A and Host B, as demonstrated in Figure 11.30. FIGURE 11.30 Place a hub between two hosts to troubleshoot. This method will allow you to see the frames sent from Host A to Host B. The bad news, however, is that by doing this, you'll bring down the network temporarily. The port-mirroring option allows you to place a port in span mode so that every frame from Host A is captured by both Host B and the sniffer, as shown in Figure 11.31. This would also be a helpful option to take advantage of if you were connecting an IDS or IPS to the switch as well. FIGURE 11.31 Port spanning/mirroring Be careful when using port mirroring because it can cause a lot of overhead on the switch and possibly crash your network. Because of this, it's a really good idea to use this feature at strategic times, and only for short periods if possible. The last thing I want you to bear in mind is that RSPAN extends SPAN by enabling remote monitoring of multiple switches across your network. The traffic for each RSPAN session is carried over a user-specified RSPAN VLAN, which is dedicated for a specific RSPAN session in all participating switches. Jumbo Frames Jumbo frames are just that, bigger frames! The maximum size of an Ethernet frame is 1500 bytes, and a jumbo frame is 9000 bytes. iSCSI storage benefits from these jumbo frames. If jumbo frames are used, data is less likely to be fragmented into smaller 1500- byte frames. When packet fragmentation occurs, the higher-level protocol of IP at layer 3 must reassemble the fragments. The reassembly of these fragmented packets causes latency and higher CPU utilization. Latency is the enemy of SANs. A caveat to jumbo frames is that all the network equipment in the switching path must support this larger framing of data, also called the maximum transmission unit (MTU). If one of the switches doesn't support the jumbo frames MTU and you turn on jumbo frames at each end (initiator and target), you could end up with a performance decrease of up to 30% or higher! Summary In this chapter, I talked about the differences between switches and bridges and how they both work at layer 2 and create a MAC address forward/filter table to make decisions about whether to forward or flood a frame. I also discussed problems that can occur if you have multiple links between bridges (switches) and how to solve these problems by using the Spanning Tree Protocol (STP). This chapter also introduced you to the world of virtual LANs and described how switches can use them. We talked about how VLANs break up broadcast domains in a switched internetwork—a very important, necessary thing because layer 2 switches only break up collision domains, and, by default, all switches make up one large broadcast domain. I also described access links and went over how trunked VLANs work across a Fast Ethernet link. Trunking is a crucial technology to understand well when you're dealing with a network populated by multiple switches that are running several VLANs. Exam Essentials Remember the three switch functions. Address learning, forward/filter decisions, and loop avoidance are the functions of a switch. Understand the main purpose of the Spanning Tree Protocol in a switched LAN. The main purpose of STP is to prevent switching loops in a network with redundant switched paths. Remember the states of STP. The purpose of the blocking state is to prevent the use of looped paths. A port in the listening state prepares to forward data frames without populating the MAC address table. A port in the learning state populates the MAC address table but doesn't forward data frames. A port in the forwarding state sends and receives all data frames on the bridged port. Last, a port in the disabled state is virtually nonoperational. Remember to check a switch port's VLAN assignment when plugging in a new host. If you plug a new host into a switch, then you must verify the VLAN membership of that port. If the membership is different than what is needed for that host, the host will not be able to reach the needed network services, such as a workgroup server. Understand what PoE provides. Power over Ethernet was created to provide power to devices that are connected to a switch port but that are not in a place that has a power outlet—for example, an access point in a ceiling. Understand what jumbo frames provide. Jumbo frames allow the maximum transmission unit (MTU) to be extended to 9000 bytes from the default 1500 bytes.

Use Quizgecko on...
Browser
Browser