Understanding ATM networking and network layer switching, part two
Confused by all the switching technology out there? We make sense of it all by looking at the top vendors' products and detailing how each works
The concepts of network layer switching that we discussed in last month's column finally get some meat as we dive into the specific technologies and standards surrounding it. We compare cut-through route switching and packet flow routing and examine technologies from Ipsilon, Cisco, 3Com, and others. (2,400 words)
etwork layer switching combines the versatility of fast hardware-based ATM switching networks with the power of router-based networks. Layer 3 switching, as it is also known, understands the flow of data between end points and provides the fastest service for delivery depending upon the traffic patterns. We looked at the two concepts of routing and switching last month, as well as some of the semi-political ramifications of the industry. Now we'll continue on and examine the actual technology implementations from various vendors.
There are several vendors in the Layer 3 switching business, most with their own proprietary implementations. Currently no approved interoperable standard exists for all products. Vendors, including Cisco, Ipsilon, 3Com, and Cabletron, each want the industry as a whole to accept their technologies so they can gain the greatest market share. These are all competing and mostly incompatible technologies; the end result is that many in the industry may either be locked into one vendor relationship or will have to overhaul their networking hardware to incorporate network layer switching.
We will first look into each of these vendor proposals and then examine what the industry standards groups are doing about it. Unfortunately we cannot do product performance comparisons. Not all of these implementations tackle the same problem; some are limited by scale, others by versatility. Any such direct product comparison would be limited or inconclusive.
First the technical nitty gritty
Conceptually these technologies can be grouped into two subsets: cut-through route switching and packet flow routing. Cut-through switching examines several incoming packets with the same source and destination addresses and "predicts" that that flow will continue for a short while. Based upon this assumption and programmed intelligence, it will then create an ATM SVC for that flow of data so that the packets become completely switched instead of individually routed through the network.
Packet flow routing, on the other hand, does not require ATM switching to become a core part of the network. It examines the same set of source and destination addresses and then applies an identifier indicating that the packets of this nature belong to a particular flow. The individual devices on the network then have the opportunity to look for this flow identifier and then send it along its way. Using just the flow identifier, the network devices do not have to examine the entire packet with all its options for each and every packet. This makes processing much faster and allows for quicker delivery of the packet.
One of the basic problems with all networks and switched networks in particular is one of interconnection complexity. Each additional switch that you add can require new links to all other switches -- at maximum the total number of switches that you have. This is mathematically called the N-squared problem: The addition of a new point into an N-sized network may require the addition up to N possible interconnections between all devices for the best connectivity. At that rate, you increase the costs dramatically as your network grows larger. Some of these Layer 3 switching techniques directly attempt to solve this problem in various ways.
Draft documents are being worked on by the IETF including topics such as multiprotocol label switching (MPLS) and Next Hop Routing Protocol (NHRP). The latter is an interesting mechanism that has a different way of looking at network traffic direction. Without going into the details of it, it forms the basis for several of the technologies we will look at.
Although the term "IP switching" is now used generically to describe Layer 3 switching of the Internet Protocol, it was originally a proposal put forward by a relative newcomer to the industry called Ipsilon. Founded by industry veteran Tom Lyon, Ipsilon began in August 1995 to tackle the issue of increasing the performance of network transmission by analyzing the flow of traffic. Its technology was awarded the Grand Winner of the Spring 1996 Interop show.
Ipsilon's technology is strengthened by two main protocol components known as the Ipsilon Flow Management Protocol (IFMP) and the General Switch Management Protocol (GSMP), given the IETF informational RFC numbers 1953 and 1987, respectively. Keep this in mind: These are NOT Internet standards! No matter what the salespeople or engineers will tell you, this statement is true. It is important that you recognize that these are informational RFCs providing an open description of how the protocols work; they are not technology standards chosen by the IETF, just open informational documents of proprietary standards.
IFMP provides a standard for inter-device communication in a network. It is not limited just to switches. What it does is allow a network device to examine the flow of traffic between a source and destination address, and then passes specific instructions to the next device along the path.
An ATM switch along the path that supports IFMP also needs to support routing to make best use. Either on its own or by examining IFMP packets, it can change the internal traffic flow of that data path from a packet-by-packet routing basis into an ATM virtual circuit that simply lets all the packets go straight through without having to decipher each packet every single time. In other words, IFMP allows devices to perform cut-through switching on an individual device basis or even across an entire WAN.
GSMP provides a means for IP networks to manage and control ATM switches. The ATM Forum already has provisions for ATM-level management of switches across the network using OAM cells and the PNNI standard. What GSMP allows is for switches to communicate across an IP network. This means that you don't have to have an all-switched network. You can have switches and routers combined into the network architecture. GSMP will take switch control information at the IP packet level and then pass it on to other switches through any number of routers. All this allows switches to communicate better and provides improved performance through shared intelligence.
Cisco's approach to the integration of Layer 3 switching into its diverse networking technologies is through what it calls tag switching. Essentially this is a flow-based routing system whereby each packet is tagged with an identifier and passed along the network. A set of tables mapping tags to source and destinations is maintained across the network, and all routers and switches keep their own set of live flows passing through them.
Cisco actually supports two types of performance enhancing switching technologies -- tag switching and NetFlow. NetFlow collects statistics on packet traffic on your network. It also allows you to monitor and filter specific traffic flows according to your network policies. It is a companion technology to tag switching and provides some of the features that the former does not. On the other hand, it can be argued that these technologies compete in some of the services that they provide.
The routing of information is based upon the tag in each packet that is looked up from the table before it is passed on. It is similar to routing but skips past the need to process all the options per packet repeatedly during its journey. It can even work with non-tag switched networks because the IP packets still essentially remain the same.
A tag switched network is made up of tag edge routers and tag switches. The overall system utilizes a tag management protocol called the Tag Distribution Protocol (TDP) and a packet forwarding technology called Cisco Express Forwarding (CEF).
A packet begins life at the source wherein it is given a data payload, a destination address, and probably some packet options. It is sent to the first router that may or may not be part of the tag network. The first actual router that supports tag switching is called the tag edge router. This router examines the packet and creates a tag and an entry in its tag information base (TIB). The edge router then generates a TDP packet describing this newly generated tag and sends it along to the next route along the path. Until the TDP packet is received the data packets are delivered and routed the old fashioned way. When a TDP packet is in hand, a new entry is created in its TIB, and the route is checked; this device then sends a TDP packet to the next hop along the flow. This way a whole tag switched route is created. Once in place, the devices only look for tags contained within the packet rather than deciphering the entire packet.
Cisco claims that this will allow a very diverse network architecture consisting of not just a switched ATM network but also of routers, bridges, and other gateways independent of the vendor.
10 licensees have agreed to incorporate this technology into their own networking systems: Adaptec (which also supports Ipsilon's technologies), AG Communications Networks, Efficient Networks, Net2Net, Network General, NUKO Information Systems, Olicom, Optical Data Systems, Radcom Limited, and Whitetree. Most are medium to small network companies. Whitetree, now part of routing and modem giant Ascend, may change its tune depending on Ascend's overlordship. Microsoft's licensing of Cisco's Internetworking Operating System means it will also include tag switching in a future version of Windows NT.
3Com's attitude with FastIP is not one for enabling faster routing and switching but rather adds intelligence to the whole concept. Essentially, FastIP introduces what 3Com calls a "third dimension" to speed and distance of network traffic, namely, policy.
FastIP is admittedly a LAN topology and traffic enhancement. At each client and server machine, FastIP is installed into the network and protocol stacks. Using a protocol known as distributed Next Hop Routing Protocol (dNHRP), the two systems can investigate if there are Layer 2 alternatives to Layer 3 connections and can make use of them if they exist. dNHRP is much like the IETF's concept with the difference lying in the fact that it does not need a route server system; instead discovery of hops and routing is left to each FastIP station.
FastIP works well within the LAN but will break down when the number of routes increases as in a WAN situation. In fact, 3Com advises against this. To provide the interconnect in a WAN, 3Com has worked in concert with the scalable systems of Cascade's IP Navigator technology.
Ascend/Cascade's IP Navigator
Ascend, continuing on its buying spree, has acquired Cascade Communications, one of the leaders in the frame relay and ATM switching markets on July 1 this year. It consequently acquired Cascade's IP Navigator technology.
IP Navigator is a technique for WANs. What is does is essentially build the IP routing software directly into the switch, eliminating the need for an additional router. It relies on the Open Shortest Path First (OSPF) routing protocol to send link state announcements amongst other switches of the same kind. This enables it to determine the best route across the network as well as establish quality of service parameters for the given circuit across the network.
The Ascend/Cascade frame relay and ATM switch products also support point-to-multipoint connections, which allows one device to connect to multiple devices as a single virtual circuit. This reduces that mathematical complexity from an N-squared problem to an N-level problem, indirectly reducing costs as well.
ATM Forum's MPOA
Multiprotocol over ATM (MPOA) was created by the ATM Forum to support Layer 3 switching within the foreign language of this technology. MPOA allows the network layer protocols to make use of the quality of service and other functional features available natively in ATM. Unlike basic classical IP over ATM that essentially just uses an ATM network as a collection of fast, fat data pipes, MPOA can work with resource reservation protocol (RSVP) parameters to ensure quality of service delivery.
The downside of MPOA is that unlike the technologies we described earlier, which were developed by individual vendors and later distributed to groups, MPOA suffers from painfully slow collaborative standards development. Also, it is more complex than most in that it attempts to provide a general Layer 3 switching system for any network protocol -- IP, IPX, etc. Thus, MPOA is still under investigation after a few years of work. The current prediction is that it will be widely available before the end of the first quarter of 1998.
Also unlike the other technologies, MPOA is only for ATM switches. MPOA gives a switch the routing capabilities of other networking systems. It is an evolution of the ATM Forum's LAN emulation (LANE) services standard for Layer 2 bridging in an ATM network. MPOA shares roots in LANE, classical IP over ATM, and the upcoming NHRP standard as well. It is based on groups of MPOA servers (route servers, configuration servers, default forwarding servers) and MPOA clients (hosts, edge switches, routers, etc.) called an Internet Address Sub-group (IASG). Several IASGs form an MPOA domain. The various servers perform the mapping of network layer addresses to ATM addresses and route the circuit between the appropriate clients.
Keep in mind that this is not a packet-based network, so the look-up of addresses from a route server is not a continual barrage of requests for every packet of data. MPOA does not necessarily provide for increased performance of data delivery; it is a standard to allow full utilization of an ATM network by upper layers and is very thorough about it. Provisions for data delivery speed-up are left up to the vendor. The MPOA stack is considered huge by some vendors because of all the features that it provides; vendors with competing proprietary protocols and methods point that out whenever they can and say that this will result in poor performance. This, however, remains to be seen; the size of the stack code may be exhaustive, but when it is done properly it can be a performance gazelle as well.
Which one is it?
It's unlikely that all these standards will exist in five years. Many fill the same niches and are therefore directly competitive; others patch those holes that are not filled appropriately. The hope is that a general Internet standard will provide a unified solution, but the realities are much harsher. The ATM Forum will continue with its MPOA standard and so will the IETF. Both overlap in some areas and are mostly independent implementations of similar concepts. On the other hand, the big name vendors like Cisco and 3Com who already have a very large share of their particular markets will try everything to keep their territory using their own technology. There are no truly overriding reasons to choose any one of these technologies over the other; it depends on your network configuration. If you have a heavily router-based network, you might opt for Cisco's method, whereas you might might choose MPOA if your network is ATM-switch centric. Unfortunately, we cannot have good news all the time. We will report back when the situation changes.
About the author
Rawn Shah is vice president of RTD Systems & Networking, Inc. He has worked with many different aspects of the LAN world and is currently strongly investigating the world of ATM and DSL and their implementation and implication on the future of voice, video, Internet, and data networking. Reach Rawn at firstname.lastname@example.org.
If you have technical problems with this magazine, contact email@example.com