Datacenter networks have become more complex over the last several years. As customer demand for enhanced services such as firewalls, virtual private networks, and load balancing increases, so does the need for better network planning and management. Multilayer switching can help datacenters maintain better network control, improve performance, and save money, without the need for additional bandwidth.

Multilayer switching has been around for several years and continues to grow in popularity. In fact, some switching companies, such as Santa Clara, California-based Extreme Networks (www.extremenetworks.com), provide multilayer functionality for all of their switching products.

Out with the old
Before delving into multilayer switching, let us first take a look at a traditional network. In a traditional network, a server is plugged into a switch, which is connected to a router. The switch is a Layer 2 device that forwards packets based on media access control (MAC) addresses. It does very little intelligent processing. The router handles the forwarding of packets to and from the server using Layer 3 protocols, such as border gateway protocol (BGP) and open-shortest-path-first (OSPF).

Unfortunately, there are several problems with this model. Scalability is the biggest problem. Although this type of two-tiered network model can be scaled, it takes an inordinate amount of network planning.

According to Ian Foo, a technical marketing engineer with Cisco (www.cisco.com; San Jose, California), "Layer 2 can indeed scale to great sizes when carefully designed, but because of the lack of fault isolation, it tends to become more complex and less robust in doing so."

Another problem is the cost. Routers have a per-port cost that is significantly higher than the per-port cost of switches. Routers also have to perform their route calculations within the software, which has a high cost in terms of CPU and memory usage and creates a network bottleneck. This is especially evident when implementing complex routing solutions.

Finally, this model tends to result in unused resources. A traditional Layer 2 switch does not load-balance traffic. Although Layer 2 switches can be set up for redundancy, they use the spanning tree protocol (STP) to block traffic to inactive ports. This prevents broadcast storms within a network, but it also means that the second device is sitting unused, unless there is a problem.

In with the new
Multilayer switching attempts to resolve many of the problems associated with a two-tier network design. Generally, a multilayer switch consists of a switch chassis with a router card installed. When a packet is received at the Layer 2 level, it is forwarded along the switching fabric to the route processor. The router determines where to forward the switch and sends the information back along the switching plane. The packet is encapsulated in a frame with the router's MAC address as the source, the Time To Live field is counted down in the IP (Internet protocol) header, the frame check sequence is recalculated, and the packet is forwarded on to its destination. This method of routing is sometimes called "route once, switch many."

Querying and updating the routing tables creates the biggest performance drain on the router. Ordinarily, a router has to perform this query for every packet it routes. In the case of a multilayer switch, the route processor does the lookup once and passes the information to the switch, which forwards future packets accordingly, without having to query the router again.

"You can essentially look at multilayer switches as integrated, feature-rich, high-performance routing," Foo says.

Even when queries have to be made to the routing tables, there is still a performance increase because, unlike a traditional router, routing decisions within a multilayer switch are made in the application-specific integrated circuit (ASIC). By moving the routing decisions from the software level to the hardware level, there is a significant increase in performance. Depending on the type of routing protocols a hosting provider is running, performance gains of 20 to 100 times are not uncommon.

Bill Robbins, the vice president of networking at BlueGenesis (www.bluegenesis.com), says BlueGenesis has been using multilayer switching since their datacenter opened.

"It provides Layer 3 at wire speeds, using ASICs instead of traditional CPUs, which allows us to save resources on our core routers by completing tasks on the switch that were previously handled by the routers," says Robbins. "They also provide superior performance at a price point much less than equivalent routers."

This performance increase, combined with the lower per-port cost of a multilayer switch, is a strong reason to migrate to a datacenter environment built around multilayer switching.

"Depending on the switch's specific task ... a performance increase and cost decrease were experienced," Robbins says, citing cost and performance as reasons for using multilayer switching.

Network necessities
In order to take advantage of the capabilities of multilayer switching, a hosting provider has to use routing protocols within the datacenter and have multiple paths for data to travel. The two most common routing protocols used in conjunction with multilayer switching are OSPF and BGP. The protocols serve different purposes and are used in different parts of the datacenter.

OSPF is an interior gateway protocol (IGP) introduced as a replacement to the routing information protocol (RIP). An IGP is a routing protocol that operates within an autonomous system or network. OSPF has several advantages over other IGPs. OSPF is able to:

• propagate changes quickly;
• send updates when there are changes in the network state;
• can be easily segmented to control the routing table size; and
• is an open standard and hierarchical.