Delivering Broadband Access Today: An Access Technologies Primer
In this chapter, you learn about the following topics:
Technologies and Protocols for DSL, Cable, and Ethernet Broadband Networks
Bridged and PPP Access Mechanisms, with an Evaluation of how well They Solve the Requirements for Broadband, Such as Quality of service, Address Assignment, Service Selection, and so on
A VPN is a service that can carry pure data or multiservice traffic. When you design or implement a VPN for broadband access, you need to understand how the different access architectures can impact design decisions and can actually have some interesting repercussions on the VPN service itself, because of quality of service (QoS) or security trade-offs, to name but two examples.
This chapter reviews the two principal Layer 2 access architectures in use today: bridging and PPP. It looks at how each is implemented on different Layer 1 broadband media, such as digital subscriber line (DSL) and cable. Then it describes how each architecture solves some of the basic requirements of a network service, such as security, QoS support, routing, and address assignment. There are lots of permutations to go through: For example, security on a bridged cable broadband network is different from security on a bridged Ethernet broadband network, so it is worthwhile to look at each case.
If you sometimes feel lost going through all of these different scenarios, remind yourself that you are looking at how common problems are solved on different types of broadband networks. The set of problems is important, because you will want to make sure that a broadband VPN service solves it too. The major topics that are covered are as follows:
Bridged access architectures
Bridging on DSL using Routed Bridge Encapsulation (RBE), including setup, routing, and address assignment
Bridging on cable
Bridging on Ethernet
Security for bridged access, with a look at different scenarios for DSL, cable, and Ethernet
Authentication and accounting for bridged access
PPP access architectures
PPP over Ethernet (PPPoE), including setup, routing, and address assignment
PPP over ATM (PPPoA)
PPP address assignment
PPP authentication, accounting, and security
Bear in mind that this is a review chapter. If you are comfortable with PPP and bridging, then you can safely skip ahead to the next chapter.
Architecture 1: Bridged Access Networks
Bridged access networks are so named because they transport Ethernet frames transparently across a network. Ethernet is the most successful LAN protocol ever. It has basically replaced all other forms of Layer 2 encapsulation in enterprise networks and is arguably in the process of doing the same thing in residential networks. Not so very long ago, subscribers connected to the Internet directly from their PC using a modem. Today, home networks use Ethernet. For example, laptops have built-in Gigabit Ethernet ports, and wireless LAN is very quickly proving to be an alternative to running cables between rooms the world over. All these scenarios use Ethernet framing, and the most cost-effective broadband service will be Ethernet-centric: Ethernet ports are cheaper, Ethernet cards are cheaper, and Ethernet equipment is cheaper, too. If Ethernet is the user-to-network interface of choice, broadband access networks need some way to carry Ethernet traffic from the subscriber premises to their destination. The easiest way to do this is to simply bridge the trafficafter all, bridging was invented to connect Ethernet LAN segments together, so it should be a pretty useful way to carry Ethernet over WAN connections, too.
However, most broadband networks are not Ethernet based, so the Ethernet frames transmitted by a device on a home network must be converted to some other form before being carried over the native transport medium. For that very reason, bridging in a DSL environment is tricky, because today's DSL uses ATM as the modulation layer of choice. To carry Ethernet, you have to do a form of RFC 2684 bridging. Cable networks are easier because they can natively encapsulate Ethernet frames directly over Data Over Cable Service Interface Specification (DOCSIS). Of course, the simplest scenario of all is one where the access network is Ethernet based, using either standard Ethernet Layer 1 or some form of optical transport over longer distances.
The advantage of bridged architectures is their simplicity. The customer premises equipment (CPE) has no difficult tasks to perform, so it can be very cheap. The overall simplicity has one significant cost, however: namely security. Unless the router enforces some form of security mechanism, all bridged subscribers are in the same broadcast domain and everyone in that domain can see sensitive traffic such as ARP requests, Windows neighbor discovery packets, and so forth. This is a situation best avoided.
Although DSL, cable, and residential Ethernet networks each use radically different transport mechanisms, the issues and design considerations of bridged access are common across all the different media. The next section looks at the details of bridging in DSL networks. This is probably the most complicated scenario, because of the conversion back and forth between ATM cells. Fortunately, bridging over ATM is well standardized in RFC 2684, and the focus of the following discussion is on how bridging over ATM works on Cisco aggregation routers.
Bridging in DSL Using RFC 2684
RBE is a Cisco implementation of bridged Ethernet over ATM with a separate broadcast domain for every ATM circuit. The CPE is a simple bridge that encapsulates Ethernet frames into ATM cells using the RFC 2684 bridging standard. Figure 2-1 shows a typical RBE architecture.
Figure 2-1 RBE Architecture
Figure 2-2 illustrates the packet encapsulations used at different points in the network.
Figure 2-2 RBE Network Cross Section
The flow of packets in Figure 2-2 works as follows:
For upstream traffic:
The subscriber PC is configured with the aggregation router's IP address as the default gateway. Just as on any Ethernet network, the PC sends an ARP request for the router MAC address and, once it learns it, transmits the Ethernet frame.
The bridged CPE encapsulates the Ethernet frame in an AAL5 bridge protocol data unit (BPDU), then segments the BPDU into ATM cells and sends it across the DSL network. Figure 2-3 shows the protocol encapsulations used at different points of the network.
The router reassembles the ATM cells, removes the AAL5 information and the Ethernet frame information, and routes the packet to its destination. Note how the router behaves: this is the behavior you would expect to see on a routed interface.
For downstream traffic:
A server sends a packet to the subscriber PC that is routed to the aggregation router.
The aggregation router has a static route that identifies the interface to use to reach the subscriber's IP address.
If necessary, the aggregator issues an ARP request to discover the subscriber PC MAC address. Then it encapsulates the Ethernet frame in an AAL5 Bridged format BPDU, segments everything into ATM cells, and transmits it.
The CPE reassembles the ATM cells into AAL5 PDUs, removes the AAL5 information, and transmits the frame on its Ethernet port.
The PC receives the data.
Figure 2-3 Payload Format for Bridged Ethernet/802.3 PDUs (source: RFC 2684)
For neighbor-to-neighbor traffic, note that if a subscriber PC sends a packet to another subscriber connected to the same aggregation router, the flow of packets is identical to the upstream and downstream flows described here. It is important to understand that there is no direct Layer 2 path between subscribers and that all traffic must be routed, even when subscribers' circuits are terminated on the same physical port on the router.
In Cisco IOS Software terms, the router in Figure 2-2 uses a logical point-to-point subinterface for each subscriber and treats each of these interfaces as a separate IP network. The default requirement of such a topology is, of course, to have a different IP subnet on every link. But in broadband, you have to manage very large numbers of connections, and there can be thousands of RBE subscribers connected to a single router. In such a case, IP addresses can run out very quickly. To get around this, you use unnumbered interfaces.
When using unnumbered interfaces, as in Frame Relay networks, the router no longer can know which interface to use to send traffic to a particular subscriber just by looking at the destination IP address, because no IP address space is associated with a subinterface. Additionally, to save IP address space, the subscriber IP addresses belong to the same subnet. Therefore, there must be an explicit route statement that maps the subscriber virtual circuit to its IP address. This is why Step 2 for downstream traffic mentions a routebecause use of the unnumbered link.
Before learning about RBE configuration, you should understand the alternative to RBE, called integrated routing and bridging (IRB), because both RBE and IRB are used (although there is less and less use of IRB). IRB is a multipoint topology in which all the subscribers are terminated on a point-to-multipoint interface. The architectural problem with IRB is that all of the subscribers are on the same Layer 2 network and are thus part of the same broadcast domain, which makes a network open both to performance degradation because of broadcast storms and to security issues. RBE is a superior, more secure implementation. For example, ARP spoofing is not possible with RBE because an ARP request for a particular address is sent only on the subinterface for that address. With IRB, the request would be flooded to all interfaces in the bridge group. RBE also prevents MAC address spoofing, again because there is a distinct subnet for each subinterface. If a hostile user tries to hijack someone else's address by injecting a gratuitous ARP packet (using their MAC and the victim's IP address), Cisco IOS will detect a subnet mismatch and generate a "Wrong Cable" error.
Note that, from a subscriber's point of view, they both look exactly the same.
Now that you understand the theory and architecture, you are ready to look at some configuration scenarios:
Basic RBE configuration
RBE IP address assignment
These sections all get into the details of Cisco IOS commands.
RBE router configuration in Example 2-1 is straightforward.
Example 2-1 Basic RBE Configuration
interface Loopback0 ip address 192.168.1.1 255.255.255.0 no ip directed-broadcast ! interface ATM0/0/0.132 point-to-point ip unnumbered Loopback0 no ip directed-broadcast atm route-bridged ip pvc 1/32 encapsulation aal5snap ! interface ATM0/0/0.133 point-to-point ip unnumbered Loopback0 no ip directed-broadcast atm route-bridged ip pvc 1/33 encapsulation aal5snap
The configuration is very similar to the regular IP over ATM on a Cisco router, with only the addition of atm route-bridge ip to enable RBE. The subscribers' hosts must be configured to use the aggregator interface as their default gateway, in this case 192.168.1.1.
RBE Quality of Service
RBE has a full range of QoS options. Because it runs over an ATM PVC, you can fully exploit all the capabilities of the ATM layer to offer different QoS profiles to subscribers. You need to remember that ATM class of service (CoS) is applied to any and all traffic on the circuit. You can't restrict it to an individual application or destination.
Additionally, you can also use IP QoS. Cisco IOS has numerous bells and whistles that let you apply policies to combinations of application flows, IP destinations, and so forth. You can classify packets, police their rate, queue them, prioritize themwhatever it is you need to do to have different levels of service made available to user applications. You enable IP QoS by applying a service policy to a PVC. Example 2-2 adds a PREMIUM policy to output IP traffic on PVC 1/33. The PREMIUM policy is not included here, but is defined using standard Cisco IOS Modular QoS CLI (MQC) syntax.
Example 2-2 RBE with IP QoS
interface ATM0/0/0.133 point-to-point ip unnumbered Loopback0 no ip directed-broadcast atm route-bridged ip pvc 1/33 encapsulation aal5snap service-policy output PREMIUM
RBE supports Weighted Random Early Detection (WRED), low-latency queuing (LLQ), and policing.
As previously mentioned, RBE uses unnumbered point-to-point subinterfaces with a route to each subscriber IP device. Example 2-3 shows the Cisco IOS routing commands, with the required static route for each subscriber.
Example 2-3 RBE Static Routes
! network routes router ospf 100 redistribute static 192.168.14.0 0.0.0.255 area 0 !subscriber routes ip route 192.168.1.2 255.255.255.255 ATM0/0/0.132 ip route 192.168.1.3 255.255.255.255 ATM0/0/0.133
Example 2-3 has just three lines of subscriber static routes, but imagine a configuration with 10,000 subscribers connected to the same router. If you announce all these host routes across an IP backbone, you can run into trouble with route table sizes, because each individual entry in a route table consumes memory.
Figure 2-4 shows a simple network with RBE subscribers connected to an aggregation router. Consider the flow when the subscriber PC in this network pings the server at 184.108.40.206.
Figure 2-4 RBE Packet Flow
In Figure 2-4, the following happens:
The subscriber PC uses the aggregation router's address as the default gateway, so the ICMP ECHO request packet is sent to R0.
The aggregation router, R0, also has a simple default route that points to R1. R0 forwards the ICMP packet received from 192.168.1.2 to 192.168.15.1, which is the address of R1.
R1 forwards the packet out its egress interface across the Internet. Assuming that IP routing is functioning correctly, the ICMP packet will eventually reach its destination at 220.127.116.11.
Now, consider how the ICMP reply is routed back to 192.168.1.2. R1 has announced the 192.168.0.0/16 network to the Internet, so data sent to 192.168.1.2 will reach it.
R0 has announced /32 routes for all the RBE subscribers, so R1 will find a route entry for 192.168.1.2/32. (Of course, the next hop will be some intermediary router between R1 and R0.)
R1 sends the ICMP packet, which is routed to R0. R0 now looks up 192.168.1.2 in its routing table and finds the static route that points to PVC 1/32 on ATM interface 0/0/0.
Multiply this scenario by several thousand RBE interfaces on several hundred aggregation routers; throughout the 192.168.0.0 network, they quickly grow to unmanageable sizes. It is a much better design to aggregate the routes as soon as possible, and using simple a static route to Null0 is a way to do this. (Plenty of other ways exist, such as configuring OSPF to announce subnet addresses, but traffic to an interface, even Null0, is processed very quickly on a router, so performance is quite good using this method.) In the example shown in Figure 2-4, the ISP 192.168.0.0 backbone routers now need to carry only a single announcement for network 192.168.1.0/24, as indicated in Example 2-4.
Example 2-4 RBE Static Routes on R0 with Null0 Route
! subscriber routes ip route 192.168.1.2 255.255.255.255 ATM0/0/0.132 ip route 192.168.1.3 255.255.255.255 ATM0/0/0.133 ip route 192.168.1.0 255.255.255.0 Null0 ! default route ip route 0.0.0.0 0.0.0.0 ATM1/0/0.100
Now, suppose host 192.168.1.2 sends a ping to 18.104.22.168. The following would happen:
On the path from 192.168.1.2 to 22.214.171.124, everything happens as before.
On the return path, the server at 126.96.36.199 replies with an ICMP REPLY, which finds its way to R1.
R1 has a route for 192.168.1.0/24 that was originally announced by R0. The ICMP REPLY packet will be forwarded to R0.
R0 has the same static route to PVC 0/132 on ATM interface 0/0/0.
Any traffic received by R0 that is for the 192.168.1.0/24 subnet, for which there is no static RBE route, will be forwarded to Null0 (i.e., dropped).
Although the use of route aggregation is well understood in large IP networks, it has not been widely used in DSL wholesale scenarios, where traffic is tunneled, not routed, to the ISP. Route aggregation is one of the challenges that reappears with IP VPNS and will be discussed further in later chapters.
In the next section, you will see how to create the subscriber routes automatically, instead of statically as in this section. Keep in mind, however, the importance of being able to aggregate as early as possible: You don't want tons of /32 routes wandering around your network.
RBE Address Assignment
You have seen RBE subscribers in all the examples so far with addresses already configured. How did they get them? How do you scale address assignment methods for broadband networks?
Because the preceding sections are all about bridging, DHCP is the logical choice for dynamic address assignment. Statically configuring all the end-station addresses obviously is impossiblethe headaches this would create would completely outweigh any employment protection advantages for network operations staff.
When using DHCP on a DSL network, you have two basic options:
Configure a DHCP server on the aggregation router. This configuration is less common, but entirely possible. You will see configuration examples of Cisco IOS DHCP servers in the "Cable CMTS" section, later in this chapter.
Use a central DHCP server to which the aggregation router forwards DHCP requests. In this case, the router behaves as a DHCP relay agent.
To do DHCP relay, add the ip helper-address command to every subscriber interface. The ip helper-address command gives the address of the DHCP server, as shown in Example 2-5.
Example 2-5 RBE Configuration with DHCP Relay
interface Loopback0 ip address 192.168.1.1 255.255.255.0 no ip directed-broadcast ! interface ATM0/0/0.132 point-to-point ip unnumbered Loopback0 ip helper-address 192.168.2.100 no ip directed-broadcast atm route-bridged ip pvc 1/32 encapsulation aal5snap
The sequence of events when using DHCP relay is as follows:
When the subscriber host starts up, it broadcasts a DHCP Discover packet. This is carried in a BPDU to the aggregation router. The aggregation router recognizes this as the DHCP packet and knows that it needs to forward it to a DHCP server because of the ip helper-address on the subinterface. In this case, the aggregator is behaving as a DHCP relay agent.
The relay agent actually converts the DHCP broadcast into a unicast IP packet to the DHCP server located at 192.168.2.100. The relay agent puts its own address in the giaddr field of the DHCP packet and puts the subscriber VPI/VCI in the Option 82 field. (This data also includes the receiving interface name, so it is unique per device. The combination of the giaddr IP address and Option 82 yields a globally unique circuit ID.) You can use the global rbe nasip command to set the interface address the router puts in the giaddr field.
You can configure multiple DHCP servers by entering additional ip helper-address commands.
The DHCP server returns a DHCP Offer packet.
The PC chooses from the different servers that replied to its Discover message and sends a DHCP request to one of them. Remember, the PC still does not have an IP address at this point, so it broadcasts.
The DHCP relay agent again forwards the packet to the DHCP server. The relay agent does have an IP address, so it unicasts the packet to the server using UDP.
The DHCP server selects an address from an appropriate pool of IP addresses (it can use either the requesting MAC address or giaddr and Option 82 to select the scope) and returns a DHCP reply to the PC, which now has its own IP address and with it the default gateway address.
Crucially, the router dynamically creates a host route for the new IP address. As DHCP replies are sent by the DHCP server, the aggregation router looks at the address being assigned and creates a host route to that address using the interface on which the request was originally received. This is one of the bits of magic needed for large-scale deployment. These routes are marked as static in the output of the show route command.
You should still use the summarization technique discussed with dynamic addresses also. Announce in your favorite dynamic routing protocol the subnet of addresses that you know will be allocated to DHCP requests originating from a particular aggregation router. As new hosts connect to the network, host routes for them are created automatically as soon as they are assigned addresses. The aggregator will then have an aggregate route to make sure that packets are sent to it for the group of potential subscribers, and specific host routes for hosts that are actually active. This way you have the best of both worlds: An aggregate route is announced to peer routers, but per-subscriber routes are dynamically created as IP addresses are assigned.
More Bridged AccessCable and DOCSIS
The worlds of cable and DSL have some major differences, but from an IP perspective they are very similar. If you ignore the many Layer 1 details on a cable headend router, the router configuration is similar to RBE, and thus many of the points already introduced for RBE also apply to cable access.
Cable modems communicate with headend routers, called CMTS, over the HFC plant using the DOCSIS standard:
Data is modulated and demodulated using the North American DOCSIS specifications, with downstream 6-MHz channels in the 54- to 860-MHz range and upstream ranges of 5 to 42MHz. The cable interface supports NTSC channel operation, using standard (STD), Harmonic Related Carrier (HRC), or Incremental Related Carrier (IRC) frequency plans conforming to EIA-S542.
NTSC uses a 6MHz-wide modulated signal with an interlaced format of 25 frames per second and 525 lines per frame. NTSC is compatible with the Consultive Committee for International Radio (CCIR) Standard M.PAL, used in West Germany, England, Holland, Australia, and several other countries.
The DOCSIS radio frequency (RF) specification defines the RF communication paths between the CMTS and CMs (or CMs in STBs). The DOCSIS RF specification defines the physical, link, and network layer aspects of the communication interfaces. It includes specifications for power level, frequency, modulation, coding, multiplexing, and contention control.
This DOCSIS standard is extremely rich, but, at a very high level, provides a TDM-like system of time slots on a shared infrastructure. In the downstream direction, variable-length MPEG-4 frames carry Ethernet frames. In the upstream direction, fixed-length time slots are assigned to the cable modems by the CMTS. The downstream bandwidth is policed by the CMTS according to the QoS profile of each subscriber. Standard Ethernet 802 LLC is run on top of the DOCSIS layer. Figure 2-5 shows the encapsulation stack.
Figure 2-5 DOCSIS Protocol Stack
As part of the session negotiation process, a Service Identifier, or SID, which is part of the DOCSIS Cable MAC layer, is allocated to each cable modem (CM). This is used somewhat like the ATM Circuit Identifier in DSL networks. All traffic sent to and by a given cable modem uses the same SID. The DOCSIS 1.1 specification enhances this to allow cable modems to use several SIDs, each with a different QoS profile so that voice or video can be run over the same infrastructure as data traffic. Again, the parallel with ATM PVCs is apparent.
The cable modem bridges traffic from its LAN Ethernet port over the WAN DOCSIS interface to the CMTS. Subscriber hosts see a shared-access Ethernet network. For upstream traffic, they behave just as RBE clients and need to ARP for the CMTS MAC address. The CMTS also ARPs for PC MAC addresses in the downstream case. Figure 2-6 shows the encapsulations used at different points in the network.
Figure 2-6 DOCSIS Network Cross Section
DOCSIS Cisco IOS Configuration
From a Cisco IOS perspective, there are commands specific to the cable plant (HFR) interfaces, cable-modem profiles, etc. Unlike RBE, cable interfaces are natively point to multipoint, which is less secure than point to point. The CMTS and CM have other techniques. Another difference between basic cable and RBE configuration, illustrated in Example 2-6, is the widespread use of secondary addressing (which is also supported with RBE, but is not used very much). In Example 2-6, the primary subnet is for the cable modems; the secondary subnet is for the hosts.
Example 2-6 Basic Cable Router Interface Configuration2
interface Cable4/0 ip address 10.1.1.1 255.255.0.0 ip address 188.8.131.52 255.255.0.0 secondary load-interval 30 no ip directed-broadcast cable helper-address 184.108.40.206 no keepalive cable downstream annex B cable downstream modulation 64qam cable downstream interleave-depth 32 cable downstream frequency 525000000 Cable upstream 0 power-level 0 no cable upstream 0 shutdown Cable upstream 0 frequency 37008000 cable upstream 1 shutdown cable upstream 2 shutdown cable upstream 3 shutdown cable upstream 4 shutdown cable upstream 5 shutdown
Cable-modem profiles are an important component of DOCSIS networks. These profiles contain configuration instructions, such as upstream and downstream bandwidth, the number of allowed hosts per connection, etc. Example 2-7 shows four different profiles.
Example 2-7 Cable-Modem Profiles3
! cable config-file platinum.cm service-class 1 max-upstream 128 service-class 1 guaranteed-upstream 10 service-class 1 max-downstream 10000 service-class 1 max-burst 1600 cpe max 10 timestamp ! cable config-file gold.cm service-class 1 max-upstream 64 service-class 1 max-downstream 5000 service-class 1 max-burst 1600 cpe max 3 timestamp ! cable config-file silver.cm service-class 1 max-upstream 64 service-class 1 max-downstream 1000 service-class 1 max-burst 1600 cpe max 1 timestamp ! cable config-file disable.cm access-denied service-class 1 max-upstream 1 service-class 1 max-downstream 1 service-class 1 max-burst 1600 cpe max 1 timestamp
Cable Address Assignment
Given that cable broadband is a bridged environment, it shouldn't be surprising to learn that DHCP is used for address assignment. There is a small quirk, though. Even though it functions as an Ethernet bridge, the cable modem also needs an IP address so that it can be managed and it can retrieve its configuration profile. Cable standards mandate the use of Trivial File Transfer Protocol (TFTP) and TOD protocols to retrieve configuration files. TFTP needs an IP address, so the modems use DHCP to get one.
Subscriber hosts also use DHCP to get their addresses. However, for security reasons, the end stations are typically on a different IP subnet than the modems.
The DHCP function on the CMTS router is quite sophisticated. You can either configure the CMTS to relay the DHCP requests to different servers, depending on whether the modem or host sends the packet, or configure different DHCP pools on the router itselfone pool for the modems, one pool for the subscribers. You can also use a mix of the two approaches.
Because of this, the Cisco IOS commands on the CMTS are a little different from RBE and you use cable helper-address instead of the standard ip helper-address command, as demonstrated in Example 2-8.
Example 2-8 Cable Router with Multiple DHCP Relay4
interface Cable3/0 ip address 220.127.116.11 255.0.0.0 no ip directed-broadcast no keepalive cable insertion-interval 500 cable downstream annex B cable downstream modulation 64qam cable downstream interleave-depth 32 cable downstream frequency 128025000 no cable downstream if-output cable upstream 0 frequency 28000000 cable upstream 0 power-level 0 no cable upstream 0 fec no cable upstream 0 scrambler cable upstream 0 data-backoff 5 12 no cable upstream 0 shutdown cable helper-address 18.104.22.168 cable-modem cable helper-address 22.214.171.124 host
In this example, there are two DHCP servers. The 126.96.36.199 server receives DHCP requests from cable modems. The 188.8.131.52 server receives requests from subscriber hosts. If you don't specify the [host | cable-modem] parameter, all requests are forwarded to a single server.
The cable dhcp-giaddr command is another powerful addition to CMTS. It modifies the giaddr field with different relay addresses, as demonstrated in Example 2-9.
Example 2-9 Using the cable dhcp-giaddr Command
interface Cable4/0 ip address 172.16.29.1 255.255.255.224 secondary ip address 10.1.4.1 255.255.255.0 cable dhcp-giaddr policy
The policy parameter instructs the router to use the DHCP pool that matches the primary address for requests from the cable modems and to use the pool that matches the secondary address for host requests.
Broadband EthernetEthernet to the Home/Business (ETTX)
Ethernet is still the brave new world of broadband access, and many aspects of broadband Ethernet continue to evolve as market demand and technical solutions develop.
Ethernet can be used in many different service types, which can create some confusion. An Ethernet access network may offer Layer 2 and Layer 3 VPN services as well as other IP-based services, such as Internet access. Figure 2-7 shows the hierarchy of Ethernet services.
Figure 2-7 Ethernet Services
The scenario of interest here, referred to as ETTX, is when Ethernet is used as a last-mile technology for a Layer 3 service, be it Internet or VPN access. The service offering can be for small and medium-sized businesses or residential customers, but today, residential Ethernet is still confined to a metropolitan area.
Unlike DSL and cable, ETTX does not use an existing wiring plant, so it is only cost effective in places where there is an abundance of Category 5 copper cable or fiber, namely multitenant buildings or metropolitan areas. There is a lot of effort today to use Ethernet framing over copper. The next generation of high-speed DSL will, in all likelihood, be Ethernet based. Cisco had an early implementation called Long Reach Ethernet (LRE), covered in more detail in the next section.
Getting back to ETTX, a common residential Internet access architecture, as shown in Figure 2-8, uses a 24-port 10/100 switch as a CPE. Each CPE has a GE trunk port that is ultimately connected to an aggregation router. Figure 2-8 shows a typical configuration that uses switches connected together in a ring between CPE and aggregators that is used to transport Gigabit Ethernet frames.
Figure 2-8 Residential ETTX
In a typical network today, there might be 10 CPE on a ring and 20 such rings per aggregator. Although sizable, the number of subscribers per aggregator is still low compared to DSL. Of course, the connection speed is many times higher. The distance from the CPE to the end station is either the standard 100 meters for Category 5 copper cable or several kilometers when using fiber. In the second case, an additional fiber-to-copper converter is required on the customer premises.
The network operation is straightforward and is identical in many ways to a switched Ethernet network in a campus. Figure 2-9 shows the by-now-familiar cross section of protocol encapsulation across the network. Frames are switched from CPE to the aggregation router across intermediary Layer 2 devices. CPE connects to 10/100 Ethernet ports, which are typically trunked over a Gigabit Ethernet port using 802.1q VLANs. The VLANs are terminated on an aggregation router (so they are switched across the access domain), which is responsible for routing traffic to its destination. As with campus networks, each VLAN runs a different IP subnet.
Figure 2-9 802.1q Network Cross Section
Figure 2-10 802.1q Header
Long Reach Ethernet
Long Reach Ethernet (LRE) is another Ethernet-based solution, this time found within MxUs using copper wiring. Figure 2-11 shows a typical network architecture. Like DSL, LRE can use twisted-pair wiring.
Figure 2-11 LRE Solution
A typical LRE deployment scenario would be a hotel that wants to offer data and voice services to its guests. Because all hotels already have a telephone network, it makes sense to use this expensive infrastructure for data transmission rather than rewire. And because the data traffic is all Ethernet based, it makes no sense to use true DSL, which would require ATM. However, standard Ethernet transmission over telephony-grade wiring is not technically possible, and this is what LRE addresses. LRE uses a transcoding scheme that allows high-speed Ethernet at up to 15 Mbps to be offered across telephony-grade wiring. An LRE CPE encodes the Ethernet frame and transmits it to an LRE-capable switch.
LRE allows voice traffic to either be carried on the same wire using existing analog bandwidth, as with DSL, or to be migrated to IP. Note that LRE works with analog and digital telephones, even if DSL is also used across the same pair of wires. LRE is not widely deployed today as a consumer solution because of ongoing issues with signal interference. It is unclear whether this solution will remain cost effective given the ever-increasing success of wireless Ethernet.
At a basic level, the ETTX configuration is identical to a campus solution. A CPE switch runs 802.1q on a Gigabit Ethernet trunk interface, with each access port in the same VLAN. In the case of Example 2-10, the CPE uses VLAN2.
Example 2-10 ETTX CPE Configuration
! CPE access port interface GigabitEthernet0/6 switchport access vlan 2 ... ! CPE trunk port interface GigabitEthernet0/1 switchport trunk encapsulation dot1q ...
The Ethernet aggregator configuration has a trunk port and an IP subinterface for every VLAN, as demonstrated in Example 2-11.
Example 2-11 ETTX Aggregator Configuration
Interface GigabitEthernet1/0/0.22 encapsulation dot1Q 2 ip address 192.168.11.1 255.255.255.240
There are networks with a different VLAN for every subscriber, but others in which a different VLAN is used for every service. In the case of per-subscriber VLANs, there must be a scaling mechanism of some kind because the maximum number of VLANs by default is 4000. (The trick is to add a second VLAN tag, known as QinQ.) When there is a VLAN per service, all subscribers are on the same IP subnet.
ETTX Quality of Service
ETTX is often perceived to have the weakest QoS infrastructure of the three access network types under consideration. Although there is no standardized equivalent to ATM's classes (CBR, VBR, etc.) or DOCSIS, Ethernet switches do offer relatively rich QoS capabilities, such as IP- and TCP-based classification, IP DSCP or 802.1p tag-based prioritization, and sophisticated scheduling and policing. Additionally, COS-to-IP DSCP mapping can be done automatically, or COS can be set on a port basis depending on the trust that is ascribed to a user.
Cisco IOS access lists can be used for classification, so different applications or hosts can be treated differently. Even quite low-cost switches can offer multiple queues per port, which is required for multiservice applications.
As anecdotal evidence, remember that a lot of enterprises run voice over their switched infrastructure, which is a testimony to the level of QoS that Ethernet infrastructure can provide.
Apart from transporting multiservice traffic, IP QoS can also be used to help compensate for the fact that Ethernet does not offer many increments for service offerings, with jumps from 10, 100, and 1000 Mbps. Subinterfaces can be policed to lower or intermediate rates, such as 2 Mbps, 34 Mbps, and others, as demonstrated in Example 2-12. You can mark down nonconforming packets, or discard them, to enforce the particular service contract.
Example 2-12 ETTX QoS ConfigurationPolicing Subscriber Interfaces
policy-map option-128k class class-default police 128000 10000 10000 conform-action set-prec-transmit 0 exceed-action drop policy-map option-512k class class-default police 512000 10000 10000 conform-action set-prec-transmit 0 exceed-action drop policy-map option-1Meg class class-default police 1000000 10000 10000 conform-action set-prec-transmit 0 exceed-action drop policy-map option-10Meg class class-default police 10000000 10000 10000 conform-action set-prec-transmit 0 exceed-action drop interface GigabitEthernet1/0/0.15 desc VLAN connecting to customer1 encapsulation dot1Q 15 ip address x.x.x.x y.y.y.y service-policy input option-128k service-policy output option-128k Interface GigabitEthernet1/0/0.88 desc VLAN connecting to customer2 encapsulation dot1Q 88 ip address a.a.a.a b.b.b.b service-policy input option-1Meg service-policy output option-1Meg
ETTX Address Assignment
Unsurprisingly, ETTX uses DHCP for address assignment to simplify address management and distribution. If, as is common today, the network is owned and operated by a single entity, there are no new issues related to address assignment beyond those already discussed thus far in the chapter. Addresses are assigned using DHCP, and the Ethernet CPE in Figure 2-8 adds option 82 information if port identification is required. The role of the CPE is very important for security, as discussed in the next section.
Using IP addresses efficiently is just as important in ETTX networks as in any DSL or cable network. Consider the case in Figure 2-12 of an Open Access ETTX network in which each subscriber can belong to one of two ISPs, ISP A or ISP B, both of which use DHCP to assign addresses.
Figure 2-12 Open Access Architecture for Residential Ethernet
Ethernet can be delivered over point-to-point or ring topologies. Although Figure 2-12 shows just one ring, there can be multiple rings of CPE connected to every aggregation router. Each ring is terminated on a physical interface, with potentially many subinterfaces, each one corresponding to a different VLAN (and there are different policies for how VLANs are used, as previously discussed). As each VLAN corresponds to a different IP subnet, there must be as many subnets as there are VLANs. For an ISP, this can result in wasted address space.
To understand the issue of address waste, consider the following sequence of steps:
ISP A has 60 subscribers in the metropolitan region and wishes to use the 192.168.1.0/26 subnet.
Subscriber A connects on the first ring and sends a DHCP request, which is relayed to ISP A's DHCP server. VLAN20 is used.
The second subscriber connects, but this time in a different part of the city and on a different ring. This time, traffic is in VLAN34.
ISP A would like to have the same subnet for all the subscribers in this metro area, but needs a different subnet for VLAN34, or else the aggregation router could not route traffic correctly to subscribers on different subinterfaces.
The bad solution to this problem is to use a different subnet for every VLAN. Unfortunately, ISP A probably has no way of knowing how many subscribers will be on each VLAN and so would potentially need to use as many /26 subnets as there are VLANs in the network. This results in huge waste.
The solution is for the metro service provider to use unnumbered interfaces. Each subscriber has a loopback interface configured for his pool and all the VLANs are unnumbered, as demonstrated in Example 2-13.
Example 2-13 Unnumbered Interfaces and Loopbacks for ETTX
interface loopback 0 desc ISP A ip address 192.168.1.1 255.255.255.0 interface loopback 1 desc ISP B ip address 192.168.2.1 255.255.255.0 interface GigabitEthernet1/0/0.15 encapsulation dot1Q 15 ip address unnumbered loopback 1 ip address secondary unnumbered loopback 2 Interface GigabitEthernet1/0/0.88 encapsulation dot1Q 88 ip address unnumbered loopback 1 ip address secondary unnumbered loopback 2
Open access is already widely deployed for DSL, but it is still a relatively new concept for ETTX networks. The architecture will continue to evolve. In Example 2-14, the addresses used by the different ISPs can't overlap, because they are terminated on the same router. You will see how to lift this restriction in Chapter 7, "Implementing Network-Based Access VPNs Without MPLS."
Security Considerations for Bridged Broadband Architectures
Security is an important part of an Internet access service, whether it is sold to residential or business customers. Security at the transport layer and application layers is beyond the scope of this work and, indeed, is independent of the type of access used. However, lower-layer security is an important part of overall network design. If the lower layers are not secure, then it is easy for an attacker to work up the stack and compromise application data such as usernames, passwords, credit card numbers, and so on.
The common Layer 2 and Layer 3 risks are as follows:
Address spoofingThis category loosely encompasses all attempts to modify an end station address. It can be something as simple as changing the address on a Linux station NIC or manually changing your IP address.
ARP-based attacks are more sophisticated and, because ARP is not an inherently secure protocol, spoofing is not as difficult as it should be. These attacks involve, for example, sending an ARP reply packet with a spoofed IP address. Most routers or switches simply overwrite an IP address in their ARP table with the IP address obtained from the most recent ARP response, or may record multiple IP addresses from responses, making it easier for the attacker. Slightly more sophisticated is the use of gratuitous ARP, whereby the attacker spontaneously advertises an ARP packet with its MAC address and someone else's IP address. This is completely RFC compliant and all stations that receive the ARP packet happily install the spurious MAC/IP mapping in their ARP tables. Now, no ARP request will be sent when traffic is received for the IP destination. The use of gratuitous ARP is typically used for man-in-the-middle attacks.
As you can gather from the preceding explanation, ARP attacks are limited only to stations that are on the same broadcast domain as the offending attacker.
DoS attacksDoS attacks can be mounted in many ways. Some simple examples include using gratuitous ARPs to fill the CAM table on an Ethernet switch; using DHCP requests to exhaust IP addresses; or sending a very large number of Layer 3 flows to the default router. DoS attacks often use some form of address spoofing. DoS attacks can be very hard to prevent. Simply constantly changing Ethernet source addresses can be very effective against a switch.
Sophisticated techniques are under development to improve network security against DoS attacks. Today's routers already have source address checking, access lists, and NetFlow statistics.
Broadcast traffic and OS weaknessesThis is not really a category of network attacks, but more an observation that many host stations are inherently insecure "out of the box" and allow any neighbor machine on the same broadcast domain to browse disk contents. In remote access, the ability to broadcast is trouble.
Security in DSL Broadband Networks
RBE has two characteristics that contribute to network layer security:
The router uses point-to-point interfaces.
The broadcast domain is limited to a single site.
If you send a gratuitous ARP packet, the router may install it in its ARP table, but downstream traffic is always directed to the correct PVC because of the host routes used with RBE. When the router receives a packet for a host, it sends the ARP only on the PVC that matches the host route for that address. In other words, ARP is used only after the correct subscriber interface has been identified by the IP routing table. Layer 2 attacks are hard to do successfully in this case.
Even if you do successfully spoof an IP address, the host routes again make sure traffic reaches the correct destination, as described in the following step sequence:
Host A sends a packet with a spoofed address of 184.108.40.206 to 192.168.1.100. Its true address is 192.168.1.99.
The RBE router forwards the packet to 192.168.1.100.
On the return path, the router either routes the traffic to the correct interface for address 220.127.116.11 or drops it if no such address exists in the routing table. It is not returned to host A.
Before you conclude that this technique is a good way to send large and unsolicited streams of traffic to your neighbor, remember Unicast Reverse Path Forwarding (uRPF) checking. uRPF, if configured, will drop the packet because a router will not accept a packet if the incoming interface is different from the interface defined in the route table that reaches the packet's source address.
Security in Cable Broadband Networks
Because security has long been an issue on cable networks, the Baseline Privacy Interface was added to the DOCSIS specifications to give a secure communication channel between each cable modem and the CMTS.
Unlike RBE, the CMTS router uses a point-to-multipoint interface. Potentially, then, ARP-based attacks are possible because the CMTS sends ARP packets to all hosts on its physical interface. However, remember that the DOCSIS layer also offers protection. The cable modem can store the MAC address/SID mappings, which the network administrator can poll to troubleshoot security issues.
The cable source-verify command is important. It configures the CMTS to enforce address assignment by dropping packets with source addresses that it has not seen assigned by a DHCP server (using the cable source-verify dhcp option). For this to work properly, the DHCP server must support the DHCPLEASEQUERY message. The cable source-verify command prevents attacks based on theft of IP addresses (using a valid address that belongs to someone else) as well as attacks based on invented addresses (either addresses that are valid but not assigned or addresses that are made up).
Security in Ethernet Broadband Networks
Without VLANs, all subscribers in a switched network are in the same broadcast domain, which is an open invitation to trouble, because of the risks that this scenario creates (some of which were discussed earlier in this chapter). Dedicating a VLAN to each customer is, in theory, possible, but it is impractical because of the 4096 global VLAN limit in the 802.1q protocol. It is more common to configure a single VLAN per switch. This still leaves everyone on the same switch in the same broadcast domain. To solve this issue, private VLANs are used.
Private VLANs prevent traffic from a subscriber port from going to any other port on the switch, with the exception of the trunk port, which is defined as a promiscuous port. From the perspective of the subscriber, as represented in Figure 2-13, subscribers "see" a private point-to-point link to the router. The aggregating router, in turn, sees a switched Ethernet segment with multiple subscribers.
Another alternative is to use double VLAN encapsulation, called QinQ, where traffic from each subscriber's port is mapped to a different 802.1q tag, and then that frame is tagged again when it leaves the switch with a tag that uniquely identifies the switch on the Ethernet network. The aggregation router has to be smart enough to handle this double layer of VLAN tags.
Figure 2-13 Private VLAN
Limiting the broadcast domain stops some of the simplest attacks, but does not prevent ARP or IP address spoofing. Remember that for cable, the router enforces DHCP assignments. If a host tries to change an address, or sends a gratuitous ARP packet, the CMTS ignores it, because it was not assigned to the host by a known DHCP server. The Ethernet scenario is harder to manage because subscriber interfaces have been aggregated on the switch downstream in the network, which may not yet have the necessary mechanisms to enforce Layer 3 to Layer 2 bindings.
Port security is a useful feature that prevents the switch from allowing a MAC address learned on one port to be used on another. Port security can also allow static MAC addresses to be configured for each port. Admittedly, it is hard to scale a solution based on static addresses for a large number of subscribers without an excellent OSS system.
Ethernet switches do have an increasingly large array of tools to deal with DoS attacks. These tools include broadcast suppression, ARP throttling, route processor rate limiting, security ACLs, and so on. On good-quality switches, these features are all implemented in hardware, so they are very efficient and are well worth some extra cost.
Identification and authorization should always be done as soon as possible in a remote-access network. There is a huge difference in effectiveness between being able to apply policies on a device where each subscriber is on a different port or Layer 2 circuit and one where this is not the case.
Currently, the 802.1x standard is emerging as a possible solution to this problem. 802.1x, which is a port-based access mechanism, works as follows:
The client station, or supplicant, sends a request to the switch, or authenticator, which forwards it to a RADIUS authentication server.
The authentication server returns a challenge to the supplicant, which must correctly respond to be granted access to the network.
The authenticator provides access to the LAN.
This is reminiscent of Point-to-Point Protocol (PPP) authentication and you can think of 802.1x as using a PPP-like authentication mechanism to provide port-based access control.
Authentication and Accounting in Bridged Broadband Architectures
One of the significant differences between Ethernet-based access and the PPP-based solutions is how subscribers are authenticated on the network.
With bridging, there is no authentication using a subscriber's name. This is always ideal, if you have the option, because you authenticate an individual and can then enforce policy with a fine level of granularity.
In bridged architectures, the network-access control is Layer 2 based. If you have a valid MAC address, from a valid port, you will receive a valid IP address. At no time does the user have to enter a name and password. So the user identity must always be tied back to the Layer 3 or Layer 2 addresses and, as you've just seen, these are not the most secure.
Billing is another weakness of the Ethernet solution if the service provider wants to offer a metered service. There is no standards-based way to retrieve usage statistics. Some switches have some useful data in MIBs, but some don't. On routers, if there is hardware for it, you can enable NetFlow accounting. However, NetFlow accounting on high-speed networks generates a considerable amount of data, and the OSS systems must do a lot of data crunching and cross checking between systems to work out which flow belonged to which person at a given time.