Is It a Switch or Is It a Router?
Keith Turner
Layer-3 Switching
The popularity of the Internet and corporate intranets has created a dramatic shift in the flow of traffic across networks. Today, network traffic is constantly crossing subnet boundaries as users access remote applications and servers. Routers are becoming overloaded as they attempt to manage this increased subnet traffic. As a result, vendors are pushing layer-3 switching as a means to marry the intelligence of routers with the speed and economy of switches.
Layer 3 is shorthand for the network layer of the OSI (Open Systems Interconnection) reference model. At this layer, routers perform their functions based on address information used in network protocols such as IP and IPX. Switches operate at layer 2, the data-link layer, forwarding packets based on the physical addressing used by the network media. By adding some level of layer-3 knowledge to their products, switch vendors are creating what they call layer-3 switches.
At first glance, it looks as though every vendor is approaching this technology differently. But when you get down to the nuts and bolts of each implementation, three approaches to layer-3 switching have emerged: routing switches, flow switching, and switched routing.
Routing Switches
A routing switch functions much like a conventional router, examining the layer-3 information in each packet to forward it along the appropriate path. The routing switch achieves lower cost and higher performance by reducing the number of supported functions and by moving as much of the logic into silicon as possible. In a true routing switch, route processing is tightly integrated into the switch so that packets do not need to leave the switching fabric during processing.
Routing switches will feel familiar to network administrators, as the switches behave much like traditional routers and support common routing protocols. Their feature sets must be examined closely, however, as some are often sacrificed for speed and cost. The features that are most commonly given up are support for non-IP protocols (such as AppleTalk and IPX), complex routing protocols (such as IP Multicast and OSPF), and security (such as encryption and firewalls).
Ascend, Bay Networks, Cisco, Extreme, Foundry, IBM, Intel, and Madge have all announced routing switch products. Each product offers a drop-in solution to off-load tasks from overworked routers in existing networks. In most cases, the switches can coexist with available equipment. Routing switches are primarily targeted at the campus, although at least one vendor's product, Ascend's GRF IP Switch, is targeted at ISPs and carrier services.
Vendors are also offering some innovations that will ease some of the administrative burden associated with managing routers in their routing switches. For example, the SwitchNode from Bay Networks can run in IP Autolearn mode, which allows the switch to learn the network topology by monitoring Address Resolution Protocol (ARP) traffic. ARP is used by network devices to map layer-3 to layer-2 addresses. So, in a simple topology, the Switch- Node can be dropped into the network without configuring it to run a routing protocol and without modifying the setup of any existing routers.
Flow Switching
The basic concept behind flow switching is the identification of long-lived data flows between two IP nodes. When a flow is detected by layer-3 software, a switched connection is established between the endpoints, and the flow is then forwarded by hardware at layer 2. File transfers and Web graphics are examples of types of traffic that would be identified as flows. Traffic that does not meet the flow criteria is routed normally through the network. The flow-switching concept lends itself most naturally to ATM or frame-relay environments, where flows can be mapped to virtual circuits or paths. The target deployment for flow switching is the ISP or enterprise backbone.
The two key players in this arena are the ATM Forum and Ipsilon. The ATM Forum has recently ratified a standard for flow switching over ATM called Multiprotocol Over ATM (MPOA), but this standard was slow to develop. Ipsilon took advantage of the delayed release of MPOA and delivered its own form of flow switching called IP switching. Ipsilon has developed a line of IP switches that use the Ipsilon Flow Management Protocol (IFMP) to exchange flow information. Ipsilon is drumming up support for its platform and encouraging other vendors to support IFMP in their switching and routing devices.
Switched Routing
The final approach and most difficult to nail down involves schemes to reduce the overhead of routing so that switches can perform layer-3 forwarding functions without the need for complex route calculations.
Cisco's Tag Switching architecture is a good example of switched routing. To implement tag switching, Cisco routers are software-upgraded to become either tag-edge routers or tag switches depending on their location in the network. A tag-edge router is a true router that sits on the network edge and adds addressing information, in the form of fixed-length identifiers called tags, to packets entering the network. A tag switch is a router or switch that sits on the interior network and uses the tags to determine the appropriate path through the network for each packet. The use of tags reduces the complexity of packet decoding and table lookups when forwarding packets. Cisco has also created the Tag Distribution Protocol (TDP) to allow tag routers and switches to distribute tag information. Cisco has submitted tag switching to the Internet Engineering Task Force (IETF) for standardization.
IBM's Multiprotocol Switched Services (MSS) and 3Com's FastIP support schemes are based on route serving using the Next Hop Resolution Protocol (NHRP). Using NHRP, a network client requests a route from a designated route server. If the route server can locate the target, a switched connection is established between the endpoints. This effectively removes the router from the data path. In order to support NHRP on a networked PC, MSS requires additional software, and FastIP requires a 3Com network adapter on the PC.
Most switched routing schemes are being developed to address bottlenecks in complex IP networks like those of ISPs and carrier services. In an attempt to deliver switched routing to the entire enterprise, not just to ISPs or the WAN, Ascend, IBM, and 3Com have agreed to share layer-3 switching technologies.
The MarketPlace
Layer-3 switching is still an emerging technology, and it's too early to tell whether one approach will receive better acceptance than another. Routing switches offer a familiar approach and the highest level of interoperability. Because these devices can be dropped into networks without a commitment to new standards or architectures, they may be adopted quickly. Flow switching and switched routing schemes currently rely on competing proprietary or emerging standards to boost packet-forwarding rates. This requires buyers to commit to a single vendor, and that could limit the appeal of flow switching and switched routing until standards emerge.
Flow switching lends itself to ATM or frame-relay environments, where flows can be mapped to virtual circuits.
Copyright (c) 1997 Ziff-Davis Inc. |
