Unraveling asynchronous transfer mode, part three
ATM switches aren't all created equal. Learn how to differentiate them and pick what suits you best
We have looked at how ATM technology works in theory and examined the different classes of ATM services. Now let's get into the actual product categories on the market, the hardware technologies involved, and the associated terms to find what is right for you. (2,500 words)
In our coverage of asynchronous transfer mode (ATM) technology, we have established what ATM is and what it can provide in the data networking world. We have also looked at the different classes of ATM service or traffic types that circuit-switched networks can provide. Where we once had only best effort delivery of data between any two points on the network, we can now have guaranteed, end-to-end delivery, fixed traffic rates, metered variable rates, and so on. The surrounding aura of ATM is quality of service. What network managers once could offer on their packet-switched networks with their fingers crossed can now be delivered with full confidence and no hesitation.
In this column we move on to practical applications of ATM. We need to familiarize ourselves with the terms and the hardware used in ATM networks first of all. Be warned that there are a great number of acronyms and abbreviations in the world of ATM. We will cover the few that we need to know and skip a whole group of others which do not directly affect us.
Since ATM provides services from the local area network all the way to wide area backbone services, there are varying levels of equipment designed for specific network situations. Typically, they range from the smallest edge device for workgroups and departmental use, all the way to core switches for nationwide and international data carriers like AT&T, Sprint, BT/MCI, etc. Each of these products fit into a different niche and unlike the situation with the different models of routers -- where it is usually just a difference in the number of packets the unit can process -- the ATM switch is much more complex. This complexity results in the low-end products having fewer services while the very high-end equipment can support almost anything and everything in the ATM standards documents.
The workgroup or edge device
At the very low end is the edge switch designed for small workgroups or departments. The name "edge" comes from the fact that this device is usually at the boundary of an ATM network and a LAN. Typically, the edge switch is also an interface between the traditional Ethernet network and the world of ATM. Although there are separate Ethernet LAN switches on the market, edge switches are different in that they tend to use ATM cells within their internal backbone instead of Ethernet frames.
Unfortunately the concept of the edge switch versus the LAN switch gets mired in technicality in that some vendors have merged these devices with their traditional LAN switches and often have dual-frame and cell-switching internal backbones to provide better performance for both types of service.
The edge switch can also provide hybrid 25-megabit-per-second (Mbps), copper-based ATM for the LAN. Conceptually it is still on the "edge" between copper and fiber circuitry so the name still applies well. The ATM cells and communications system are essentially still identical with the difference being only in the physical signaling systems.
There are many examples of this, especially considering that it is the cheapest to manufacture. The 3Com LinkSwitch 2700, Ascend/Whitetree WS3000, and FORE ES 3810 are all good examples of this category.
The enterprise backbone switch
The next step up from the immediate LAN is the corporate or enterprise network. In some cases enterprise backbone switches are similar to edge devices in that they may provide some Ethernet or Fast Ethernet switching capabilities. However, it has become more the case that this switch is all fiber and interconnects individual edge switches through 155-Mbps OC-3c cabling within the limitation of a corporate campus. This limitation is really in the type of fiber used -- multimode fiber.
There are currently two types of fiber in wide use today: single-mode and multimode fiber. The major differences in the two are in the diameter of the fiber cables and the distances they support. Single mode fiber is approximately eight to 10 micrometers in diameter and has a distance rating of around 20 to 40 kilometers, depending upon the quality of the cable. Multimode fiber ranges between 50 to 125 micrometers in diameter and works up to two kilometers. The smaller size, manufacturing process, and circuitry needed to support single-mode fiber makes it a more expensive option. The different data rates for fiber optic cables depend upon the type of light-emitting diode or laser used to transmit. As you go higher, more and more powerful transmission systems are needed, to the point of, for example, Lucent Technologies's new 1450 Dense Wavelength Division Multiplexers that will transmit up to 20 gigabits per second (Gbps) per fiber cable in the near future.
The enterprise switch typically supports multimode fiber and is therefore limited in the distances it will support. This keeps the price down in the range of typical routers on the network today. The switch may also have a single-mode fiber option as an uplink to a data carrier or WAN service provider. The total number of ATM ports range between eight and 16 in most cases. These support T-1 (1.544 Mbps) at the low end and go up to 155-Mbps OC-3c in most cases.
The Cisco Catalyst 5000 series falls into both the edge and enterprise switch categories depending on the model and card types you install. The Ipsilon IP Switch ATM1600 is a "special case" enterprise switch because of a built-in feature we will discuss in a coming issue. FORE's ASX-200 and the Nortel Meridian switches are other good examples.
The carrier switch
The carrier switch is typically a larger model of the enterprise switch. Additionally the carrier switch supports single-mode fiber on most if not all its ports. The backplane or cross connect between each of the ports describes the total data capacity that can be transmitted by the switch. It typically has a higher backplane bandwidth capacity than the enterprise switch, running between five and 20 Gbps instead of one to five Gbps. Furthermore carrier switches support many more ports than enterprise switches -- up to 100 or so. Many carrier switches support OC-12 speeds of 622 Mbps at the high end and T-3/DS-3 (approx. 45 Mbps) at the low end.
The 3Com Cellplex/CoreBuilder 7000 and the FORE ASX-1000 are prime contenders in this market. The Cisco (formerly Stratacom Inc.) BPX family and Cisco (formerly LightStream Inc.) Lightstream 1010 fall into this range and so does the Nortel Vector switch.
The core switch
Core switches are the gargantuans of the ATM switch family. These monsters can switch up to 1,200 T-3/DS-3 (approx. 45 Mbps) connections at a time. Think about it. At one time, the highest-speed connection across the entire Internet was at DS-3 speeds. Nowadays it is the minimum speed that nationwide service providers need to have. The essential difference between core and carrier switches is simply capacity. The AT&T/Lucent Technologies GlobalView 2000, for example, can provide this many connections and has multiple backplanes for each group of connections that can handle up to 80 Gbps of traffic for the backplane.
There's often confusion between carrier and core switches because some vendors like to identify their products with the largest and the best. Also this line is sometimes confused by souped-up versions of carrier switches which tread into this last category as well. This latter is becoming more the case these days.
If you need to ask how much a core switch actually costs, you probably don't need it. These are only used by the largest wide area network service providers of the world. At this level, the switch takes up several racks of space and can fill a small room. Typically installations are custom designed for the client and come with very good support contracts and reassurances from the vendor.
The physical link -- SONET
Synchronous optical network (SONET) forms the standard for fiber optic communications in much of ATM. Although ATM itself is independent of the raw signaling system involved, it was first designed with SONET links in mind. We will skip over the physical signaling methodology and go to practical SONET use.
SONET links are labeled as synchronous transport signal (STS) levels. The basic unit is STS1 that runs at 51.84 Mbps. The different levels run in increments of this basic rate; e.g., STS3 runs at 155.52 Mbps. In Europe, a different basic unit was devised called synchronous digital hierarchy/synchronous transfer mode (SDH/STM) at 155.52 Mbps; essentially the same STS3 but not physically compatible.
More people are familiar with the generic name of a unit of optical cabling, labeled optical carrier (OC). An STS3 or SDH/STM-1 uses an OC-3c line. The basic unit of OC is OC-1 at 51.84 Mbps. They are typically graduated in multiples of four. The OC-3c is the basic line deployed in most ATM networks; for example, my company, ATMnet, has deployed OC-3c (155 Mbps) lines in California like so many others. The next step up is OC-12c or 622 Mbps (or 4 x 155 Mbps); MCI has several OC-12c lines across the country for all the different services of MCI. GTE recently purchased part of the Qwest network across the U.S. that will work at OC-48c rates (2.4 Gbps). The maximum wide area SONET link currently implemented is OC-192c (9.6 Gbps) and only in a test environment.
Physically these lines may look the same; the difference lies in the communications hardware you use. We have already talked about the difference between single-mode and multimode fiber in ranges and costs of physical lines. As you might expect, the cost essentially goes up as the circuit gets larger.
It has been argued whether or not quirky copper-based, twisted-pair lines (the same kind used for Ethernet and telephones) can realistically provide the high speeds necessary for ATM in a robust manner. However, the availability of copper-based lines on such a wide scale globally made vendors find a way. With standard twisted pair, copper-based ATM is only possible in the LAN.
Desktop ATM25 is one standard for lower-speed connectivity for the desktop market. It is half the speed of STS1 or 25.6 Mbps and has the same range as 10BaseT Ethernet cables. The network interface cards for this standard are cheaper and are about $500. It works over Category 3 cabling at minimum.
It is possible to get STS3 (155 Mbps) speeds over Category 5 (the highest rating) lines but in a limited environment. The range varies, but a theoretical 100 meters may be possible.
Vendors such as Adaptec, IBM, Fore, and 3Com have products in this market in a price range between $400 and $1500.
PVC and SVC capabilities
All ATM switches can build permanent virtual circuits (PVCs), which we defined in our previous issue. In addition, more vendors are also incorporating switched virtual circuit (SVC) technology into their products. A PVC is simpler to operate because it is usually manually set up and dedicated to its use; you may need to account for the traffic on a PVC, but most opt for a flat-rate schedule.
SVCs require more intelligent circuitry to set up, operate, monitor, account, and tear down a link between two or more points. At any point of time, an SVC might start up and terminate after a few minutes. This means that the connection between end points lasts for very short periods of time and during that time, you have to measure how much time is used. From the user's point of view, SVCs may save a lot of money in the long run because you pay for only what you use; unless, of course, you use the connection so much, in which case you should get a PVC instead.
Within a corporation, the accountability of SVCs would depend on internal policies and departmental billing; in fact, a corporation may even opt to ignore the accounting of individual SVCs as they are set up and torn down. However, if you are a network service provider charging customers for connections, you need to accurately meter how circuits operate.
The UNI and the NNI
There are two acronyms you will come across a great deal in the ATM world -- user network interface (UNI) and network node interface or network-to-network interface (NNI). These describe the point on the network where two logical entities come in contact.
The UNI is a definition of how an end terminal like a desktop PC or a workstation communicates with the rest of the ATM network. The UNI establishes calls and connects applications to the network. Essentially it establishes a standard that all ATM cards must comply with for compatibility.
The NNI is a definition of how two separate ATM networks interface with each other and how they communicate their needs to one another. Private network node interface (PNNI) is the actual standard name for interfacing the ATM switches of different vendors.
When looking at ATM hardware, you must make sure that the proper versions of these standards are implemented. Some products are still playing catch up to others, so you should be careful.
The current standard for UNI is version 3.1; there are still implementations with version 3.0 built in but this older version is incompatible with the current one. UNI 4.0 is compatible with 3.1 and includes support for the newer Available Bit Rate service class and better multicast services. PNNI Phase I is still the current standard that not only defines how signaling between two switches works but also how to route ATM cells across the ATM network (note that this isn't necessarily the same as IP traffic routing). To support SVCs, switches must also implement the interim interswitch signaling protocol (IISP) which defines how two switches propagate SVC setup, monitoring, and tear-down procedures.
On to the virtual network
Now that we have covered some of the products and some of the associated terminology we can move on to one of the more interesting designs that can be made using ATM technology -- the virtual network. The virtual network or virtual LAN uses the point-to-point or point-to-multipoint capabilities as well as the high speeds of ATM to create networks of computers that seem to be physically connected together in the same segment. In truth, a virtual network may actually have computers in Rome, London, New York, and San Francisco all at the same time. The computers, however, do not know the difference. The benefits of such virtual networks lie in efficiency of data traffic as well as added security of information flow.
Next month we will take a look at Virtual LANs and networks as well as a new development in ATM technology which is making a lot of noise in the world of Internet routing.
About the author
Rawn Shah is director of intelligence at ATMnet, a provider of integrated digital communications services. Reach Rawn at Rawn.Shah@sunworld.com.
If you have technical problems with this magazine, contact email@example.com