This chapter covers the following topics:
L2 Switching Basics
Metro Ethernet Services Concepts
Example of an L2 Metro Ethernet Service
Challenges with All-Ethernet Metro Networks
As discussed in Chapter 1, "Introduction to Data in the Metro," Ethernet services can take either of two forms: a retail service that competes with traditional T1/E1 private-line services, or a wholesale service where a carrier sells a big Ethernet transport pipe to another, smaller service provider. In either case, multiple customers share the same metro infrastructure and equipment. For TDM deployments, sharing the infrastructure is a nonissue, because the services are limited to selling transport pipes, and each customer is allocated a circuit that isolates its traffic from other customers. The customer gets well-defined SLAs, mainly dictated by the circuit that is purchased.
When packet multiplexing and switching are applied, such as in the cases of switched EOS, Ethernet Transport, and RPR, things change. Packets from different customers are multiplexed over the same pipe, and the bandwidth is shared. No physical boundaries separate one customer's traffic from another's, only logical boundaries. Separation of customer traffic and packet queuing techniques have to be used to ensure QoS. Multiple functions have to be well-defined to offer a service:
How to identify different customers' traffic over a shared pipe or shared network
How to identify and enforce the service given to a particular customer
How to allocate certain bandwidth to a specific customer
How to "transparently" move customers' traffic between different locations, such as in the case of transparent LAN services
How to scale the number of customers
How to deploy a VPN service that offers any-to-any connectivity for the same customer
This chapter starts by discussing the basics of L2 Ethernet switching to familiarize you with Ethernet-switching concepts. Then it discusses the different metro Ethernet service concepts as introduced by the Metro Ethernet Forum (MEF).
L2 Switching Basics
L2 switching allows packets to be switched in the network based on their Media Access Control (MAC) address. When a packet arrives at the switch, the switch checks the packet's destination MAC address and, if known, sends the packet to the output port from which it learned the destination MAC.
The two fundamental elements in Ethernet L2 switching are the MAC address and the virtual LAN (VLAN). In the same way that IP routing references stations on the networks via an L3 IP address, Ethernet L2 switching references end stations via the MAC address. However, unlike IP, in which IP addresses are assigned by administrators and can be reused in different private networks, MAC addresses are supposed to be unique, because they are indicative of the hardware itself. Thus, MAC addresses should not be assigned by the network administrator. (Of course, in some cases the MAC addresses can be overwritten or duplicated, but this is not the norm.)
Ethernet is a broadcast medium. Without the concept of VLANs, a broadcast sent by a station on the LAN is sent to all physical segments of the switched LAN. The VLAN concept allows the segmentation of the LAN into logical entities, and traffic is localized within those logical entities. For example, a university campus can be allocated multiple VLANsone dedicated for faculty, one dedicated for students, and the third dedicated for visitors. Broadcast traffic within each of these VLANs is isolated to that VLAN.
Figure 3-1 shows the concept of an Ethernet LAN using a hub (Part A) and an Ethernet switch (Part B). With an Ethernet hub, all stations on the LAN share the same physical segment. A 10-Mbps hub, for example, allows broadcast and unicast traffic between the stations that share the 10-Mbps bandwidth. The switched LAN on the right allows each segment a 100-Mbps connection (for this example), and it segments the LAN into two logical domains, VLAN 10 and VLAN 20. The concept of VLANs is independent of the stations themselves. The VLAN is an allocation by the switch. In this example, ports 1 and 2 are allocated to VLAN 10, while ports 3 and 4 are allocated to VLAN 20. When stations A1 and A2 send traffic, the switch tags the traffic with the VLAN assigned to the interface and makes the switching decisions based on that VLAN number. The result is that traffic within a VLAN is isolated from traffic within other VLANs.
Ethernet switching includes the following basic concepts:
Using broadcast and multicast
Expanding the network with trunks
The need for the Spanning Tree Protocol (STP)
MAC learning allows the Ethernet switch to learn the MAC addresses of the stations in the network to identify on which port to send the traffic. LAN switches normally keep a MAC learning table (or a bridge table) and a VLAN table. The MAC learning table associates the MACs/VLANs with a given port, and the VLAN table associates the port with a VLAN. In Figure 3-1, Part B, the switch has learned the MAC addresses of stations A1, A2, B1, and B2
on ports 1, 2, 4, and 3, respectively. It also shows that ports 1 and 2 are associated with VLAN 10 and ports 3 and 4 are associated with VLAN 20.
Figure 3-1 Ethernet MACs and VLANs
If the switch receives a packet with a destination MAC address that does not exist in the bridge table, the switch sends that packet over all its interfaces that belong to the same VLAN assigned to the interface where the packet came in from. The switch does not flood the frame out the port that generated the original frame. This mechanism is called flooding. It allows the fast delivery of packets to their destinations even before all MAC addresses have been learned by all switches in the network. The drawback of flooding is that it consumes switch and network resources that otherwise wouldn't have been used if the switch had already learned which port to send the packet to.
VLANs minimize the effect of flooding because they concentrate the flooding within a particular VLAN. The switch uses the VLAN table to map the VLAN number of the port on which the packet arrived to a list of ports that the packet is flooded on.
Using Broadcast and Multicast
Broadcast is used to enable clients to discover resources that are advertised by servers. When a server advertises its services to its clients, it sends broadcast messages to MAC address FFFF FFFF FFFF, which means "all stations." End clients listen to the broadcast and pick up only the broadcasts they are interested in, to minimize their CPU usage. With multicast, a subset of broadcast, a station sends traffic only to a group of stations and not to all stations. Broadcast and multicast addresses are treated as unknown destinations and are flooded over all ports within a VLAN. Some higher-layer protocols such IGMP snooping help mitigate the flooding of IP multicast packets over an L2 switched network by identifying which set of ports a packet is to be flooded on.
Expanding the Network with Trunks
So far you have seen the case of a single L2 switch. An L2 Ethernet-switched network would consist of many interconnected switches with trunk ports. The trunk ports are similar to the access ports used to connect end stations; however, they have the added task of carrying traffic coming in from many VLANs in the network. This scenario is shown in Figure 3-2. Trunk ports could connect Ethernet switches built by different vendorshence the need for standardization for VLAN tagging mechanisms.
Figure 3-2 Trunk Ports
In Figure 3-2, switches SW1 and SW3 have assigned access port 1 with VLAN 10 and access port 2 with VLAN 20. Port 3 is a trunk port that is used to connect to other switches in the network. Note that SW2 in the middle has no access ports and is used only to interconnect trunk ports. You can see that the simplicity of switched Ethernet becomes extremely complex becauseVLAN assignments need to be tracked inside the network to allow the right traffic to be switched on the right ports. In Frame Relay, ATM, and MPLS, similar complexities do exist, and signaling is introduced to solve the network connectivity issues. Ethernet has not defined a signaling protocol. The only mechanisms that Ethernet networks have are third-party applications that surf the network and make it easier to do some VLAN allocations. While these mechanisms work in small enterprise environments, they immediately became showstoppers in larger enterprise deployments and carrier networks. Chapter 4 discusses LDP as a signaling mechanism for delivering Ethernet services. Chapter 7 discusses RSVP-TE and its use in relation to scaling the Ethernet services.
IEEE 802.1Q defines how an Ethernet frame gets tagged with a VLAN ID. The VLAN ID is assigned by the switch and not the end station. The switch assigns a VLAN number to a port, and every packet received on that port gets allocated that VLAN ID. The Ethernet switches switch packets between the same VLANs. Traffic between different VLANs is sent to a routing function within the switch itself (if the switch supports L3 forwarding) or an external router. Figure 3-3 shows how the VLAN tags get inserted inside the untagged VLAN packet.
Figure 3-3 VLAN Tagged Packet
The untagged Ethernet packet consists of the destination MAC address and source MAC address, a Type field, and the data. The 802.1Q tag header gets inserted between the source MAC address and the Type field. It consists of a 2-byte Type field and a 2-byte Tag Control Info field. The 2-byte Type field is set to 0X8100 to indicate an 802.1Q tagged packet. The 2-byte Tag Control Info field consists of the 3 leftmost bits indicating the 802.1P priority and the 12 rightmost bits indicating the VLAN tag ID. The 802.1P field gives the Ethernet packet up to eight different priority levels that can be used to offer different levels of service within the network. The 12-bit VLAN ID field allows the assignment of up to 4096 (212) VLAN numbers to distinguish the different VLAN tagged packets.
Metro Ethernet applications require extensions to L2 switching that are not defined in the standards. An example is the ability to do VLAN stackingthat is, to do multiple VLAN tagging to the same Ethernet packet and create a stack of VLAN IDs. Different entities can do L2 switching on the different levels of the VLAN stack. Cisco Systems calls this concept Q-in-Q, short for 802.1Q-in-802.1Q, as shown in Figure 3-4.
As shown, an already tagged frame can be tagged again to create a hierarchy. The simplicity of Ethernet, the lack of standardization for many such extensions, the reliance on STP, and the explosion of MAC addresses contribute to the lack of confidence of many providers in deploying a large-scale, all-Ethernet network.
Figure 3-4 Q-in-Q
VLAN tag support is discussed more in the section "VLAN Tag Support Attribute."
The Need for the Spanning Tree Protocol
L2 Ethernet-switched networks work on the basis of MAC address learning and flooding. If multiple paths exist to the same destination, and the packet has an unknown destination, packet flooding might cause the packet to be sent back to the original switch that put it on the network, causing a broadcast storm. STP prevents loops in the network by blocking redundant paths and ensuring that only one active path exists between every two switches in the network. STP uses bridge protocol data units (BPDUs), which are control packets that travel in the network and identify which path, and hence ports, need to be blocked.
The next section covers in detail the Ethernet services concepts as defined by the Metro Ethernet Forum.