7.1 Gigabit Ethernet Switches

Gigabit Ethernet switches form the core infrastructure of nearly all enterprise IP networks. The term switch is somewhat misleading, because typically Gigabit Ethernet switches also perform layer 3 routing of IP packets. The term router still connotes the high processing latency common to first-generation IP routers, whereas switch implies high performance. Current Gigabit Ethernet switches are now both layer 2 switches and layer 3 routers, with some products also performing layer 4 through 7 functions.

Gigabit Ethernet switches are marketed as both departmental 8- or 16-port products and data center-class 200+-port products. High port count data center Gigabit Ethernet switches may include high-availability features such as hot-swappable modules, nondisruptive microcode load, failover and redundant processors, power supplies, and fans. These larger Gigabit Ethernet switches are therefore equivalent to Fibre Channel directors in terms of high availability, although the per-port cost is typically half (or less) that of Fibre Channel products.

As with Fibre Channel switches, Gigabit Ethernet switches achieve high port density by using small form factor connectors. Unlike Fibre Channel switches, Gigabit Ethernet switches also support Category 5 cabling and so can provide even high density through use of standard RJ-45 ports. This enables a Gigabit Ethernet director-class switch to support more than 200 ports in a single chassis.

Switch architecture is typically based on a modular design that allows for a variety of port and processor cards to be installed in the chassis. Vendors then have the opportunity to sell varying densities of ports as well as value-added functionality supported by dedicated processors. In addition to port cards (blades) that provide 1Gbps ports, the vendor may supply blades for 10Mbps/100Mbps connections to workstations or 10Gbps modules for interswitch links, as shown in Figure 7-1. With multiple link speed modules in the same chassis, the blades must be designed with sufficient buffering to avoid flooding slower ports when they receive packets from higher-performance ports.

Figure 7-1. Gigabit Ethernet switch with 10Mbps/100Mbps, 1Gbps, and 10Gbps port modules

As additional port blades are installed, the aggregate bandwidth required for switching increases. In some vendor implementations, a fully loaded switch may result in a blocking architecture if all ports are simultaneously active. The data sheet may not reflect this fact, but you should always query the vendor on the tested performance of the switch at maximum load. Because a switch might support different types of traffic through VLAN segregation, it is important to avoid disruption of storage data as additional modules and load are added to the chassis over time.

Gigabit Ethernet switches are cut-through switches. In a cut-through implementation, only the destination address makes a switching decision. For layer 2 switching on the same network segment, only the MAC address is read. For layer 3 switching, the destination IP address is parsed from the IP header. In either case, as soon as the intended destination is known, the rest of the packet is automatically switched to the appropriate port. A store-and-forward switch, by contrast, must buffer the entire packet before a switching decision is made. IP storage switches are necessarily store-and-forward switches, because not only must the entire packet be buffered before switching, but also the contents of the packet itself undergo protocol conversion. The challenge for IP storage switch design is to do this at wire speed so that the switch appears to have cut-through performance.

The arrival of 10Gbps interswitch link modules in mid-2001 enabled Gigabit Ethernet switches to more easily scale to support thousands of devices. Like 1Gbps links, 10Gbps ports can be link-aggregated to provide fatter pipes between switches. Depending on traffic requirements, interswitch links can be provisioned to create a high-performance multiswitch network core, with fan-out to edge switches to provide departmental connectivity.

Because Gigabit Ethernet switches support layer 2 and layer 3 functionality, they must support both IEEE layer 2 standards and IETF IP routing standards. Layer 2 features include 802.3z compliance, 802.1Q VLAN tagging, 802.1p/Q frame prioritization, 802.3x flow control, and 802.3ad link aggregation. Layer 3 IP features include a long laundry list of IETF Requests for Comments (RFCs), including OSPF routing, RIP, and Border Gateway Protocol (BGP). In addition, vendors may offer auxiliary modules for MPLS and IP security, along with advanced support for management and auditing.

Using Gigabit Ethernet switches for IP storage eliminates the requirement for dedicated director-class storage switches to build data center SANs. From the standpoint of the Gigabit Ethernet core, storage data is only one more variant of IP traffic. Storage-specific functions are provided by iFCP or iSCSI gateways or iSCSI end devices. Customers can elect to devote one or more Gigabit Ethernet switches to storage, or they can create a VLAN for existing Gigabit Ethernet switches to support both storage and messaging traffic. If the network is properly designed and if sufficient attention is paid to application requirements, you can conveniently support block storage data on Gigabit Ethernet switching while taking advantage of layer 2 and layer 3 enhanced services.

Because data center Gigabit Ethernet switches are already built on a modular architecture, you can further integrate IP storage in the mainstream network by incorporating storage services modules in the switch chassis. In addition to simplifying deployment, integration centralizes management of IP storage assets and facilitates support of storage-specific operations such as virtualization.