IP Switching:

The Intelligence of Routing, the Performance of Switching

Ipsilon Networks
February 1996 

Executive Summary

LAN switching has become the solution of choice for segmenting congested client/server networks to handle the expanding bandwidth requirements of growing user populations and feature-rich applications. High speed technologies such as ATM, offering 155 Mbps throughputs today and scaling to 622 Mbps and beyond as networks evolve, have added to the appeal of switching. 

But these solutions have created new problems for existing internetworks. Switches greatly expand the number of total LAN segments while restoring the flattened topologies of bridged environments, thereby multiplying the opportunities for broadcast storms. Conventional routers cannot keep up with the increased traffic capacity of high speed switches, creating new network bottlenecks. And ATM cannot accommodate the most widely used network protocol-the Internet Protocol (IP)-without complicated software translation. 

Ipsilon Networks has developed a new category of product that solves these problems. Ipsilon's IP Switch combines the intelligence and control of IP routing with the high speed and capacity of ATM switching hardware to advance the state of the art in internetworking price/performance. Based on scalable, robust, and proven technologies, IP switches provide millions of IP packets-per-second (PPS) throughput while maintaining full compatibility with existing IP networks, applications, and network management tools. 

This paper examines the climate that has created the need for IP Switches. It then explains how an IP Switch dynamically chooses between IP routing and ATM switching, depending on the characteristics of the network traffic. The paper looks at the elegant Ipsilon system software that implements an IP Switch and describes how the Ipsilon approach delivers an order of magnitude higher throughput than is available with other solutions. It explores Ipsilon's intuitive Web-based network configuration and management interface. Finally, it presents two tactical migration scenarios for introducing IP Switches into existing internetworks. 

The Need for Switching

Since the early 1980s, when LANs began proliferating in campus environments, network managers have faced a recurring problem-how to minimize congestion and optimize throughput. The first solution was to break up large LANs into multiple smaller LANs serving smaller populations. Bridges connected the segments to form a single LAN while segregating traffic local to each segment. 

The focus on internetworking in the mid-1980s spawned the need for routers, capable of connecting different network types as well as over the WAN. Routers added structure to networks. The filtering and firewalling capability of routers provided more control over broadcast domains, limiting broadcast traffic and enhancing security. Added intelligence in routers allowed them to choose the best path through the network and to build redundant paths to destinations when possible. Of course, this extra sophistication added to the port cost of routers, and it incurred substantial latency overhead. 

Today, distributed client/server data traffic, expanded user populations, and more complex applications create new bandwidth bottlenecks for shared-media networks. Unpredictable network response times, higher network failure rates, and the inability to support delay-sensitive applications are some of the problems resulting from this congestion. 

LAN switches are rapidly being deployed as low cost, easy-to-use solutions for increasing throughput and relieving today's workgroup network bottlenecks. Switches enable fine-grained network segmentation and can deliver dedicated bandwidth per segment. Like bridges, switches make simple forwarding decisions based on link-layer addresses contained in each packet. Thus, switches have very low latencies, typically one-tenth those of fast routers. 

The low port cost and implementation simplicity of Ethernet switches, offering dedicated 10 Mbps per port, have paved the way for a migration to higher speed technologies, such as asynchronous transfer mode (ATM), Frame Relay, Fast Ethernet, and fiber distributed data interface (FDDI). ATM is emerging as the frontrunner among other high speed alternatives for several reasons: 

New Solutions, New Problems

While LAN switching provides finer segmentation and affordable high bandwidth to increase network throughput, it creates new problems in its wake. In particular, switches flatten the network hierarchy, causing switched networks to suffer from the same problems as traditional bridged networks, most notably broadcast storms and poor security. Some would say that switched networks are just a broadcast storm waiting to happen. So while virtual LANs allow switched networks to be logically subdivided, they still require routing in the backbone. 

Most routers are too slow to keep pace with the increased throughput of today's high speed switches. To fill a full-duplex 155 Mbps link requires forwarding rates at 100,000 packets per second through a single interface. Conventional routers, which compute data paths on a packet-by-packet basis, fall short of this performance goal and create new data roadblocks. The unpredictable latency of current routing technology also is a bad match for delay-sensitive traffic enabled by switching, such as desktop video and other multimedia applications. In reaction to these limitations, internetworking vendors have announced a host of new routing solutions, including edge routing, split routing using route servers, and distributed routing using multilayer switches. Most of these proprietary schemes combine routing, LAN switching, and ATM. 

LAN Switching, Return of the Flat Earth slide 

Switches taking advantage of high speed, high capacity ATM have yet another problem: interoperability with existing internetwork devices and applications. ATM is a connection-oriented technology that puts a new twist on the age-old telecom mainstay, the circuit. It exchanges fixed-length cells between two ATM stations over a virtual channel connection (a virtual circuit) using new signaling and routing protocols. 

Yet most of today's data networking protocols have been designed to operate using connectionless transmission technology based on global addressing. Herein lies the dilemma. The success of ATM as a networking technology hinges on its ability to support the protocols used by existing networks and applications, particularly IP, the dominant network-layer protocol. Proposed solutions for this interoperability problem, including Classical IP over ATM (IETF RFCs 1577 and 1483), LAN Emulation (ATM Forum LANE), and Multiprotocol over ATM (ATM Forum MPOA), have become steeped in complexity, resulting not only in delayed implementations but increasing market confusion and frustration. As a result, ATM switching has fallen short of its promises in today's networks. 

IP Switching-the Intelligent High Performance Alternative

Ipsilon Networks has developed a new class of switch that solves these problems. Ipsilon's IP Switch unites intelligent IP routing with high speed ATM switching hardware in a single scalable platform that delivers five times the performance of conventional routers at a fraction of the price. 

IP Switch-ATM Configuration 

An IP Switch implements the IP protocol stack on ATM hardware, which operates as a high performance link-layer accelerator. Using an intelligent classification scheme, an IP Switch dynamically shifts between store-and-forward and cut-through switching based on the needs of the traffic. The majority of data is switched directly by the ATM hardware, without additional IP router processing, achieving millions of PPS throughput. 

IP Switches integrate easily into existing internetworks. Based on well-understood and time-tested technology, IP Switches behave outwardly in familiar ways, making them simple to implement and to use. Routing decisions are based on IP protocols, so IP Switches behave like other IP nodes and are naturally interoperable with existing applications and network management tools. IP Switches optimize traffic in pure IP environments, or where tunneling or encapsulation of non-IP protocols (such as IPX, NetBIOS, SNA, or DECnet) is employed. 

IP Switch Attributes 

An IP Switch in Operation

The following figure shows an IP Switch in operation. 

Cut-Through Switching Summary 

At system startup, each IP node sets up a virtual channel on each of its ATM physical links to be used as the default forwarding channel. IP data traffic from existing network devices flows into an upstream host, edge router, or IP Switch gateway equipped with an ATM network interface card (NIC) and Ipsilon software. 

An ATM input port inside the IP Switch receives incoming traffic from the upstream device on the default channel and sends it to the intelligent routing software of the IP Switch Controller (1). The ATM switch hardware functions simply as a high speed I/O extension of the routing software. 

The IP Switch Controller forwards the packet in the normal manner over the default forwarding channel. It also performs flow classification, a decision-making process that enables IP Switches to optimize data traffic. A flow is an extended IP conversation. More specifically, a flow is a sequence of IP packets sent from a particular source to a particular destination sharing the same protocol type (such as UDP or TCP), type of service, and other characteristics, as determined by information in the packet header. The switch controller identifies longer duration flows, since these can be optimized by cut-through switching in the ATM hardware. The rest of the traffic continues to receive the default treatment-hop-by-hop store-and-forward routing. 

Once a flow is identified, the switch controller asks the upstream node to label that traffic using a new virtual channel (2). If the upstream node concurs, the traffic starts to flow on the new virtual channel (3). Independently, the downstream node also can ask the IP Switch Controller to set up an outgoing virtual channel for the flow(4). When the flow is isolated to a particular input channel and a particular output channel (5), the IP Switch Controller instructs the switch to make the appropriate port mapping in hardware, bypassing the routing software and its associated processing overhead (6). This design allows IP Switches to forward packets at rates limited only by the aggregate throughput of the underlying switch engine. First-generation IP Switches support up to 5.3 million PPS throughput. Further, because there is no need to reassemble ATM cells into IP packets at intermediate IP Switches, throughput remains optimized throughout the IP network. 

Independent, Local Decisions 

An important feature of the Ipsilon solution is that flow classification and switching are "soft-state" decisions local to individual IP Switches. They have no impact on the correct operation of the network. This distinction preserves the connectionless nature of IP and allows each IP Switch to maintain an independent network identity. Network state remains dynamic, allowing IP Switches to route around failures without reestablishing the circuit from the host point. 

The Ipsilon architecture also allows for flow definitions to reflect implementation requirements of each IP Switch. For example, flow traffic deep in the heart of the Internet, where IP conversations are relatively short, might have different needs from flows in enterprise LANs. The ability to change the flow definition is important to accommodate new types of traffic as they emerge. 

Support for Quality of Service 

Flow characterization allows each IP Switch to make its own quality-of-service decisions according to local priorities. IP Switches also can base quality-of-service decisions on the capabilities of the underlying ATM switch hardware. While today's switches support best-effort and real-time flows, future switch designs will offer more sophisticated scheduling capabilities. 

Individual quality-of-service requests for each flow will be supported using the resource reservation protocol (RSVP). 

IP Switching and Multicast Traffic 

IP Switches can support native IP multicast without any modification to standard IP multicast protocols, such as the distance vector multicast routing protocol (DVMRP) and the Internet group management protocol (IGMP). From the point of view of an IP Switch, multicast traffic is handled naturally by flow classification. Continuous multimedia streams, such as live video and audio from a group videoconference, map cleanly into the IP flow-classification model, allowing the ATM switch engine to handle high performance multicast replication. Short-lived multicast traffic is forwarded automatically to multiple destinations by the IP routing software. 

SIDEBAR: Flow-Classification Simulation Results 

The degree to which an IP Switch can optimize performance depends on its ability to classify the majority of traffic as flows that can be switched in hardware. To understand how much traffic qualifies as flows, Ipsilon analyzed traffic traces from the Internet backbone. The trace monitored an FDDI ring connecting users in the San Francisco Bay Area to the Internet. In this simulation, more than 80 percent of packets and 90 percent of bytes qualified as flows suitable for switching. 

Internet backbone traffic is most likely to present a worst case for flow switching because of the large number of independent conversations it represents. Thus even more traffic in campus and corporate backbone environments should qualify for the performance optimizations of an IP Switch. 

The following table distinguishes typical flow-oriented traffic from the types of data that normally do not qualify as a flow. 
Flow-Oriented Traffic  Short-Lived Traffic 
File transfer protocol (FTP) data  Domain Name Service (DNS) query 
Telnet data  Simple Mail Transfer Protocol (SMTP) data 
HyperText Transmission Protocol (HTTP) data  Network timing protocol (NTP) 
Web image downloads  Point-of-presence (POP) 
Multimedia audio/video  Simple Network Management Protocol (SNMP) querieas 

IP Switch System Software

The intelligence within the IP Switch system derives from two simple protocols: the Generic Switch Management Protocol (GSMP) and the Ipsilon Flow Management Protocol (IFMP). 

GSMP controls the ATM switch engine. It is responsible for setting up, tearing down, and monitoring the status of the virtual channels within the ATM switch fabric. IFMP enables communications between multiple IP Switches or between hosts and IP Switches. It associates IP flows with ATM virtual channels and defines the format for flow-redirect messages and acknowledgments. IFMP is implemented in end stations, such as routers, shared-media hubs, LAN switches, or TCP/IP hosts equipped with an ATM NIC to connect directly to IP Switch. 

Simplicity is the watchword for these protocols. At approximately 2000 lines of code, the GSMP software is relatively trivial; IFMP can be implemented in less than 10,000 lines of code, a small fraction of the overall IP routing protocols. In contrast, a typical ATM switch running MPOA requires about 300,000 lines of code. 

ATM Forum Protocols Versus IP Switch Protocols 

Cross-Protocol, Cross-Platform Compatibility 

The Ipsilon software is fully compatible with existing and emerging IP protocols, including the routing information protocol (RIP), open shortest path first (OSPF) protocol, distance vector multicast routing protocol (DVMRP), and the Internet group management protocol (IGMP). Support for the next-generation IP (IPng), resource reservation protocol (RSVP), border gateway protocol (BGP), and intermediate system-to-intermediate system (IS-IS) routing protocols is planned. 

In addition, the Ipsilon software is compatible with other hardware platforms beyond ATM. Because ATM offers the best price/performance for enterprise LANs, it is the logical choice for the company's flagship product line. Upcoming generations of the Ipsilon technology will address the needs to the largest consumer of IP networking-the Internet-where Frame Relay offers the best WAN price/performance today. 

Ipsilon system software specifications are available from Ipsilon's home page. To further interoperability, Ipsilon provides reference implementations and source code to ATM NIC, LAN switch, and router vendors wanting to support IP Switching directly. 

Web-Based Network Management

One of the prime advantages of the Ipsilon solution is its foundation in IP, a proven, robust, and widely understood technology. This factor alone simplifies IP Switch configuration and management. IP Switches look like a familiar devices to network management systems-like a standard IP router with many ports. So existing network management tools, techniques, and applications apply to IP Switches. Support for standard MIBs allows IP Switches to integrate naturally into the network views provided by SNMP-compatible platforms, including HP OpenView®, SunNet? Manager, IBM NetView® 6000, and Microsoft Windows?. 

But an IP Switch makes device configuration and management even easier by harnessing the most powerful graphical user interface available-the Web browser. Each IP Switch incorporates a World Wide Web server that can be accessed locally or remotely through a NetScape or other Web browser. Managers enter configuration information through intuitive, user-friendly forms, not tedious command-line interfaces. Navigating through the Web site, they can check current state of the device, refer to the IP Switch documentation, download new software versions, or link to Ipsilon customer service. 

Key IP Switch Features 

IP Switching Migration Strategies

IP Switches complement the installed based of internetworking equipment and thus can be deployed in existing networks in low risk, incremental stages. Environments with high bandwidth, low latency requirements are the most likely to benefit from the addition of IP Switching. 

A high performance workgroup is one such environment. A tactical deployment of an IP Switch in this scenario establishes an IP switched network in parallel with an existing routed network. The IP Switch provides up to 5.3 million PPS throughput to dedicated workgroup servers and power users, while the routed network handles low level and multiprotocol traffic. Legacy hosts and the routed network communicate with the IP Switch through IP Switch gateways. As the demand for IP bandwidth increases, additional users can migrate to the IP switched network and reap immediate performance gains while freeing up resources on the routed network. 

IP Switched Workgroup with Parallel Routed Network 

Building an IP switched backbone in parallel with other protocol-based backbones is another incremental migration strategy for enterprise environments. This solution delivers the speeds and performance of ATM while maintaining full compatibility with existing IP hosts and applications. A mesh topology of IP Switches connects campus buildings to the IP backbone. ATM servers can connect directly to the IP Switches to eliminate server bottlenecks, while IP gateways and routers implementing the Ipsilon protocols provide LAN connections. The existing backbone provides a fully redundant data path throughout the transition. 

IP and FDDI Parallel Backbones 

Both scenarios leverage current skills and network management tools for reduced complexity, reduced cost of ownership, and low risk migration. 

Summary

The growth path for today's internetworks must accommodate the ever-increasing demand for bandwidth from an expanding user population wanting more sophisticated services and applications. Beyond more bandwidth, the migration strategy must support existing protocols, applications, and services to preserve current network investment. It must control the cost of ownership by delivering easy-to-understand, easy-to-manage high performance solutions. And it must provide built-in scalability to accommodate future needs. The right combination of LAN switching, routing, and ATM technology can address this demand. 

Ipsilon Networks develops IP Switches, a new product category that unites the simplicity and intelligence of IP routing with high speed, high capacity ATM switching hardware to advance the state of the art in internetworking price/performance. IP Switches address the need for high bandwidth, high performance, and scalability without increasing complexity or creating new problems for existing networks. Based on scalable, robust, and proven technologies, IP Switches provide millions of IP PPS throughput while maintaining full compatibility with existing IP LANs, applications, and network management tools. 

Copyright © 1997 Ipsilon Networks, 232 Java Drive, Sunnyvale, CA 94089-1318
phone 408.990.2000, fax 408.743.5675
Last Revised: July 10, 1996