Infrastructure Device Characteristics and Considerations

Network end-user devices are commonly connected using switched technology rather than using a shared media segment. Switched technology provides dedicated network bandwidth for each device on the network. Switched networks can support network infrastructure services, such as QoS, security, and management; a shared media segment cannot support these features.

In the past, LAN switches were Layer 2-only devices. Data link layer (Layer 2) switching supports multiple simultaneous frame flows. Multilayer switching performs packet switching and several functions at Layer 3 and at higher Open Systems Interconnection (OSI) layers and can effectively replace routers in the LAN switched environment. Deciding whether to deploy pure data link layer switches or multilayer switches in the enterprise network is not a trivial decision. It requires a full understanding of the network topology and user demands.

KEY POINT

The difference between data link layer and multilayer switching is the type of information used inside the frame to determine the correct output interface.

Data link layer switching forwards frames based on data link layer information (the MAC address), whereas multilayer switching forwards frames based on network layer information (such as IP address).

Multilayer switching is hardware-based switching and routing integrated into a single platform. See the upcoming "Multilayer Switching and Cisco Express Forwarding" section for implementation details.

When deciding on the type of switch to use and the features to be deployed in a network, consider the following factors:

■ Infrastructure service capabilities: The network services that the organization requires (IP multicast, QoS, and so on).

■ Size of the network segments: How the network is segmented and how many end devices will be connected, based on traffic characteristics.

■ Convergence time: The maximum amount of time the network will be unavailable in the event of network outages.

■ Cost: The budget for the network infrastructure. Note that multilayer switches are typically more expensive than their Layer 2 counterparts; however, multilayer functionality can be obtained by adding cards and software to a modular Layer 2 switch.

The following sections examine the following infrastructure characteristics: convergence time, multilayer switching and Cisco Express Forwarding, IP multicast, QoS, and load sharing.

Convergence Time

Loop-prevention mechanisms in a Layer 2 topology cause the Spanning Tree Protocol (STP) to take between 30 and 50 seconds to converge. To eliminate STP convergence issues in the Campus Core, all the links connecting core switches should be routed links, not VLAN trunks. This also limits the broadcast and failure domains.

NOTE STP tools are covered in the section "The Cisco STP Toolkit" later in this chapter.

In the case where multilayer switching is deployed everywhere, convergence is within seconds (depending on the routing protocol implemented) because all the devices detect their connected link failure immediately and act on it promptly (sending respective routing updates).

In a mixed Layer 2 and Layer 3 environment, the convergence time depends not only on the Layer 3 factors (including routing protocol timers such as hold-time and neighbor loss detection), but also on the STP convergence.

Using multilayer switching in a structured design reduces the scope of spanning-tree domains. It is common to use a routing protocol, such as Enhanced Interior Gateway Routing Protocol (EIGRP) or Open Shortest Path First (OSPF), to handle load balancing, redundancy, and recovery in the Campus Core.

Multilayer Switching and Cisco Express Forwarding

As noted in Chapter 3, "Structuring and Modularizing the Network," in this book the term multilayer switching denotes a switch's generic capability to use information at different protocol layers as part of the switching process; the term Layer 3 switching is a synonym for multilayer switching in this context.

The use of protocol information from multiple layers in the switching process is implemented in two different ways within Cisco switches. The first way is called multilayer switching (MLS), and the second way is called Cisco Express Forwarding.

Multilayer Switching

Multilayer switching, as its name implies, allows switching to take place at different protocol layers. Switching can be performed only on Layers 2 and 3, or it can also include Layer 4. MLS is based on network flows.

KEY POINT

A network flow is a unidirectional sequence of packets between a source and a destination. Flows can be very specific. For example, a network flow can be identified by source and destination IP addresses, protocol numbers, and port numbers as well as the interface on which the packet enters the switch.

The three major components of MLS are as follows:

■ MLS Route Processor (MLS-RP): The MLS-enabled router that performs the traditional function of routing between subnets

■ MLS Switching Engine (MLS-SE): The MLS-enabled switch that can offload some of the packet-switching functionality from the MLS-RP

■ Multilayer Switching Protocol (MLSP): Used by the MLS-RP and the MLS-SE to communicate with each other

KEY POINT

MLS allows communication between two devices that are in different VLANs (on different subnets), that are connected to the same MLS-SE, and that share a common MLS-RP. The communication bypasses the MLS-RP and instead uses the MLS-SE to relay the packets, thus improving overall performance.

MLS History

Pure MLS is an older technique used on the Catalyst 5500 switches with a Route Switch Module (manually configured as the MLS-RP) and a Supervisor Engine III with a NetFlow Feature Card (manually configured as the MLS-SE). The first packet of a flow is routed by the MLS-RP, whereas the MLS-SE records (caches) all flow, or header, information; all subsequent packets in the identical flow are hardware-switched by the MLS-SE.

Most of Cisco's modern multilayer switches use Cisco Express Forwarding-based multilayer switching (as described in the next section), using hardware integrated in the switch platform.

Cisco Express Forwarding

Cisco Express Forwarding, like MLS, aims to speed the data routing and forwarding process in a network. However, the two methods use different approaches.

Cisco Express Forwarding uses two components to optimize the lookup of the information required to route packets: the Forwarding Information Base (FIB) for the Layer 3 information and the adjacency table for the Layer 2 information.

Cisco Express Forwarding creates an FIB by maintaining a copy of the forwarding information contained in the IP routing table. The information is indexed, so it is quick to search for matching entries as packets are processed. Whenever the routing table changes, the FIB is also changed so that it always contains up-to-date paths. A separate routing cache is not required.

The adjacency table contains Layer 2 frame header information, including next-hop addresses, for all FIB entries. Each FIB entry can point to multiple adjacency table entries—for example, if two paths exist between devices for load balancing.

After a packet is processed and the route is determined from the FIB, the Layer 2 next-hop and header information is retrieved from the adjacency table, and the new frame is created to encapsulate the packet.

Cisco Express Forwarding can be enabled on a router (for example, on a Cisco 7600 Series router) or on a switch with Layer 3 functionality (such as the Catalyst 6500 Series switch).

NOTE Not all Catalyst switches support Cisco Express Forwarding. See the specific product documentation on the Cisco website for device support information.

IP Multicast

A traditional IP network is not efficient when sending the same data to many locations; the data is sent in unicast packets and therefore is replicated on the network for each destination. For example, if a CEO's annual video address is sent out on a company's network for all employees to watch, the same data stream must be replicated for each employee. Obviously, this would consume many resources, including precious WAN bandwidth.

IP multicast technology enables networks to send data to a group of destinations in the most efficient way. The data is sent from the source as one stream; this single data stream travels as far as it can in the network. Devices replicate the data only if they need to send it out on multiple interfaces to reach all members of the destination group.

Multicast groups are identified by Class D IP addresses, which are in the range from 224.0.0.0 to 239.255.255.255. IP multicast involves some new protocols for network devices, including two for informing network devices which hosts require which multicast data stream and one for determining the best way to route multicast traffic. These three protocols are described in the following sections.

Internet Group Management Protocol and Cisco Group Management Protocol

Internet Group Management Protocol (IGMP) is used between hosts and their local routers. Hosts register with the router to join (and leave) specific multicast groups; the router then knows that it needs to forward the data stream destined for a specific multicast group to the registered hosts.

In a typical network, hosts are not directly connected to routers but are connected to a Layer 2 switch, which is in turn connected to the router. IGMP is a network layer (Layer 3) protocol. Consequently, Layer 2 switches do not participate in IGMP and therefore are not aware of which hosts attached to them might be part of a particular multicast group. By default, Layer 2 switches flood multicast frames to all ports (except the port from which the frame originated), which means that all multicast traffic received by a switch would be sent out on all ports, even if only one device on one port required the data stream. Cisco therefore developed Cisco Group Management Protocol (CGMP), which is used between switches and routers. The routers tell each of their directly connected switches about IGMP registrations that were received from hosts through the switch—in other words, from hosts accessible through the switch. The switch then forwards the multicast traffic only to ports that those requesting hosts are on, rather than flooding the data to all ports. Switches, including non-Cisco switches, can alternatively use IGMP snooping to eavesdrop on the IGMP messages sent between routers and hosts to learn similar information.

Figure 4-7 illustrates the interaction of these two protocols. Hosts A and D register, using IGMP, to join the multicast group to receive data from the server. The router informs both switches of these registrations using CGMP. When the router forwards the multicast data to the hosts, the switches ensure that the data goes out of only the ports on which hosts A and D are connected. The ports on which hosts B and C are connected do not receive the multicast data.

Multicast Data

Protocol-Independent Multicast Routing Protocol

Protocol-Independent Multicast (PIM) is used by routers that forward multicast packets. The protocol-independent part of the name indicates that PIM is independent of the unicast routing protocol (for example, EIGRP or OSPF) running in the network. PIM uses the normal routing table, populated by the unicast routing protocol, in its multicast routing calculations.

NOTE EIGRP, OSPF, and so forth are called unicast routing protocols because they are used to create and maintain unicast routing information in the routing table. Recall, though, that they use multicast packets (or broadcast packets in some protocols) to send their routing update traffic. Note that a variant of OSPF, called multicast OSPF, supports multicast routing; Cisco routers do not support multicast OSPF.

Unlike other routing protocols, no routing updates are sent between PIM routers.

When a router forwards a unicast packet, it looks up the destination address in its routing table and forwards the packet out of the appropriate interface. However, when forwarding a multicast packet, the router might have to forward the packet out of multiple interfaces, toward all the receiving hosts. Multicast-enabled routers use PIM to dynamically create distribution trees that control the path that IP multicast traffic takes through the network to deliver traffic to all receivers.

The following two types of distribution trees exist:

■ Source tree: A source tree is created for each source sending to each multicast group. The source tree has its root at the source and has branches through the network to the receivers.

■ Shared tree: A shared tree is a single tree that is shared between all sources for each multicast group. The shared tree has a single common root, called a rendezvous point (RP).

Multicast routers consider the source address of the multicast packet as well as the destination address, and they use the distribution tree to forward the packet away from the source and toward the destination. Forwarding multicast traffic away from the source, rather than to the receiver, is called Reverse Path Forwarding (RPF). To avoid routing loops, RPF uses the unicast routing table to determine the upstream (toward the source) and downstream (away from the source) neighbors and ensures that only one interface on the router is considered to be an incoming interface for data from a specific source. For example, data received on one router interface and forwarded out another interface can loop around the network and come back into the same router on a different interface; RPF ensures that this data is not forwarded again.

PIM operates in one of the following two modes:

■ Sparse mode: This mode uses a "pull" model to send multicast traffic. Sparse mode uses a shared tree and therefore requires an RP to be defined. Sources register with the RP. Routers along the path from active receivers that have explicitly requested to join a specific multicast group register to join that group. These routers calculate, using the unicast routing table, whether they have a better metric to the RP or to the source itself; they forward the join message to the device with the better metric.

■ Dense mode: This mode uses a "push" model that floods multicast traffic to the entire network. Dense mode uses source trees. Routers that have no need for the data (because they are not connected to receivers that want the data or to other routers that want it) request that the tree be pruned so that they no longer receive the data.

QoS Considerations in LAN Switches

A campus network transports many types of applications and data, which might include high-quality video and delay-sensitive data (such as real-time voice). Bandwidth-intensive applications enhance many business processes but might also stretch network capabilities and resources.

Networks must provide secure, predictable, measurable, and sometimes guaranteed services. Achieving the required QoS by managing delay, delay variation (jitter), bandwidth, and packet loss parameters on a network can be the key to a successful end-to-end business solution. QoS mechanisms are techniques used to manage network resources.

The assumption that a high-capacity, nonblocking switch with multigigabit backplanes never needs QoS is incorrect. Many networks or individual network elements are oversubscribed; it is easy to create scenarios in which congestion can potentially occur and that therefore require some form of QoS. The sum of the bandwidths on all ports on a switch where end devices are connected is usually greater than that of the uplink port; when the access ports are fully used, congestion on the uplink port is unavoidable. Uplinks from the Building Access layer to the Building Distribution layer, or from the Building Distribution layer to the Campus Core layer, most often require QoS. Depending on traffic flow and uplink oversubscription, bandwidth is managed with QoS mechanisms on the Building Access, Building Distribution, or even Campus Core switches.

QoS Mechanisms

QoS mechanisms or tools implemented on LAN switches include the following:

■ Classification and marking: Packet classification is the process of partitioning traffic into multiple priority levels, or classes of service. Information in the frame or packet header is inspected, and the frame's priority is determined. Marking is the process of changing the priority or class of service (CoS) setting within a frame or packet to indicate its classification.

For IEEE 802.1Q frames, the 3 user priority bits in the Tag field—commonly referred to as the 802.1p bits—are used as CoS bits. However, Layer 2 markings are not useful as end-to-end QoS indicators, because the medium often changes throughout a network (for example, from Ethernet to a Frame Relay WAN). Thus, Layer 3 markings are required to support end-to-end QoS.

For IPv4, Layer 3 marking can be done using the 8-bit type of service (ToS) field in the packet header. Originally, only the first 3 bits were used; these bits are called the IP Precedence bits. Because 3 bits can specify only eight marking values, IP precedence does not allow a granular classification of traffic. Thus, more bits are now used: the first 6 bits in the TOS field are now known as the DiffServ Code Point (DSCP) bits.

NOTE Two models exist for deploying end-to-end QoS in a network for traffic that is not suitable for best-effort service: Integrated Services (IntServ) and Differentiated Services (DiffServ). End-to-end QoS means that the network provides the level of service required by traffic throughout the entire network, from one end to the other.

With IntServ, an application requests services from the network, and the network devices confirm that they can meet the request, before any data is sent. The data from the application is considered a flow of packets.

In contrast, with DiffServ, each packet is marked as it enters the network based on the type of traffic that it contains. The network devices then use this marking to determine how to handle the packet as it travels through the network. The DSCP bits are used to implement the DiffServ model.

■ Congestion management: Queuing: Queuing separates traffic into various queues or buffers; the marking in the frame or packet can be used to determine which queue traffic goes in. A network interface is often congested (even at high speeds, transient congestion is observed); queuing techniques ensure that traffic from the critical applications is forwarded appropriately. For example, real-time applications such as VoIP and stock trading might have to be forwarded with the least latency and jitter.

■ Congestion Management: Scheduling: Scheduling is the process that determines the order in which queues are serviced.

■ Policing and shaping: Policing and shaping tools identify traffic that violates some threshold level and reduces a stream of data to a predetermined rate or level. Traffic shaping buffers the frames for a short time. Policing simply drops or lowers the priority of the frame that is out of profile.

NOTE Later chapters in this book describe two other QoS mechanisms: congestion avoidance and link efficiency techniques.

QoS in LAN Switches

When configuring QoS features, classify the specific network traffic, prioritize and mark it according to its relative importance, and use congestion management and policing and shaping techniques to provide preferential treatment. Implementing QoS in the network makes network performance more predictable and bandwidth use more effective. Figure 4-8 illustrates where the various categories of QoS may be implemented in LAN switches.

Figure 4-8 QoS in LAN Switches

Building Access Data Link Layer Switching

Building Distribution Multilayer Switching

Campus Core Multilayer Switching

Classification, Marking

Congestion Management

Building Distribution Multilayer Switching

Campus Core Multilayer Switching

Classification, Marking

Congestion Management

Congestion Management, Policing, Shaping

Data link layer switches are commonly used in the Building Access layer. Because they do not have knowledge of Layer 3 or higher information, these switches provide QoS classification and marking based only on the switch's input port or MAC address. For example, traffic from a particular host can be defined as high-priority traffic on the uplink port. Multilayer switches may be used in the Building Access layer if Layer 3 services are required.

Building Distribution layer, Campus Core layer, and Server Farm switches are typically multilayer switches and can provide QoS selectively—not only on a port basis, but also according to higher-layer parameters, such as IP addresses, port numbers, or QoS bits in the IP packet. These switches make QoS classification more selective by differentiating the traffic based on the application. QoS in distribution and core switches must be provided in both directions of traffic flow. The policing for certain traffic is usually implemented on the distribution layer switches.

Load Sharing in Layer 2 and Layer 3 Switches

Layer 2 and Layer 3 switches handle load sharing differently.

Layer 2 Load Sharing

Because Layer 2 switches are aware of only MAC addresses, they cannot perform any intelligent load sharing. In an environment characterized by multiple VLANs per access switch and more than one connection to the uplink switch, the solution is to put all uplink connections into trunks (Inter-Switch Link [ISL] or IEEE 802.1Q). Each trunk carries all VLANs; however, without additional configuration, the STP protocol disables all nonprimary uplink ports. This configuration might result in a bandwidth shortage, because the traffic for all the VLANs passes through the same link. To overcome this problem, the STP parameters must be configured to carry some VLANs across one uplink and the rest of the VLANs across the other uplink. For example, one uplink could be configured to carry the VLANs with odd numbers, whereas the other uplink would be configured to carry the VLANs with even numbers.

NOTE Some options related to STP are described in the "Building Access Layer Design Considerations" section on the next page.

Layer 3 Load Sharing

Layer 3-capable switches can perform load sharing based on IP addresses, either per packet or per destination-source IP pair.

The advantage of Layer 3 IP load sharing is that links are used more proportionately than with Layer 2 load sharing, which is based on VLANs only. For example, the traffic in one VLAN can be very heavy, while the traffic in another VLAN is very low; in this case, per-VLAN load sharing by using even and odd VLANs is not appropriate. Due to the dynamic nature of organizational applications, Layer 3 load sharing is more appropriate. Layer 3 allows for dynamic adaptation to link utilization and depends on the routing protocol design. Layer 3 switches also support Layer 2 load sharing, so they can still apply per-VLAN load sharing while connected to other Layer 2 switches.

Continue reading here: Building Access Layer Design Considerations

Was this article helpful?

+1 0