As corporate networks grow and as more network applications are deployed, companies find that they are often stressing the capacity of their networks. And with bandwidth-intensive applications (e.g., multimedia and desktop videoconferencing) on the horizon, many networks, in their current form, will not be able to handle the traffic loads. In the long term, bringing ATM (asynchronous transfer mode) to every desktop will ensure that there's enough bandwidth, but today it is far too expensive for most business networking applications.
One of the most promising and economic ways to satisfy the traffic patterns generated by newer client/server and multimedia applications is to combine two switching technologies--ATM and Ethernet switching--into one network. Such a combination gives departments more usable bandwidth, thanks to Ethernet switching and, by using an ATM backbone network, lets large amounts of data flow between workgroups (see the figure, "Merging Switching Technologies"). This combination of switching technologies can be used in a campus setting where buildings are linked through an ATM backbone, or in a single building where Ethernet switching hubs on each floor are connected through an ATM switch in the basement.
Whether used in a campus setting or within a single building, this merging of ATM and Ethernet switching technologies has many points in its favor. First, it leaves future networking options open. For example, a company could decide at a later date to push ATM to the desktop (when ATM adapter cards drop in price). It also puts a company in position to connect widely dispersed sites using high-speed, ATM WAN services that many of the telephone companies and independent service providers are gearing up to deliver.
Merging ATM and Ethernet switching technologies into one networking environment also preserves a company's current investment in its network adapter cards, cabling, and workgroup-level wiring hubs and concentrators. You can keep your existing equipment and still offer better bandwidth allocation to your departments. The cost savings of retaining the desktop are significant. For instance, an alternative to getting more bandwidth to each desktop would be to move everyone over to an FDDI (Fiber Distributed Data Interface) network. Even though prices have dropped, the least expensive FDDI adapter cards still cost several hundred dollars more than 10Base-T adapters. And such a change would also require you to replace your networking hubs.
It's not surprising then that many users are interested in merging ATM and Ethernet switching. Virtually all of the major router and enterprise hub vendors have been quick to announce that they will be players in this developing market. Some internetworking product vendors, including Cisco Systems (Menlo Park, CA) and 3Com (Santa Clara, CA), have beefed up their switching offerings through acquisitions. At the same time, hub vendors Alantec (San Jose, CA), Cabletron Systems (Rochester, NH), Chipcom (Southborough, MA), Digital Equipment (Maynard, MA), Lannet (Irvine, CA), Networth (Irving, TX), Optical Data Systems (Richardson, TX), Standard Microsystems (Hauppauge, NY), and Bay Networks (Santa Clara, CA) have been positioning their higher-end hubs and routers to enter this market.
Additionally, a handful of stand-alone Ethernet switching hub vendors have products that they've developed from the start to handle Ethernet on the workgroup side and ATM in the backbone (in the future). These products include the ANTswitch from Applied Network Technology (Westford, MA), the ATMizer 125 Relational Switch from Agile Networks (Concord, MA), the LANbooster Series from Onet Data Communication Technologies (Cambridge, MA), the MegaSwitch from NBase Switch Communications (Chatsworth, CA), and the QuikStack from XLNT Designs (San Diego, CA).
Different Approaches
There are several things to consider before selecting a device to perform Ethernet switching today with connectivity to ATM backbones in the future. First and foremost, decide which of the two fundamentally different types of product best fits your networking philosophy: An enterprise hub uses Ethernet switching and ATM modules that share an internal high-speed backplane (or ATM switching fabric within the hub), whereas a stand-alone workgroup hub performs Ethernet switching and has an interface to an ATM switch (which will be part of the ATM backbone).
The workgroup hub systems that Applied Network Technology, Agile, Onet, and NBase offer will typically be less expensive per port than will enterprise hubs. That's because, with these products, you're not paying for the features (i.e., redundant backplanes, power supplies, and cooling fans) that are required to operate in an enterprise setting. You're also not paying for the enterprise management system that is featured with enterprise hubs.
And that may be the second point to consider: How do you want to manage your network? If the workgroups are fairly autonomous, only needing connectivity to other groups, a stand-alone hub's management system will be fine. However, you will need to ensure that your hub vendor's ATM interface is compatible with the ATM switch manufacturer's product. (Yes, there are standards for such interfaces, but make sure your hub vendor has a solid technical relationship with one or two of the major ATM switch vendors.)
On the flip side, if you plan to connect users in different workgroups via enterprise-wide virtual LANs, you will need an enterprise management system. One note of caution: Even if you select equipment from an enterprise internetworking vendor, you may not get the enterprise management features you need. That's because there has been such an acquisition frenzy in this market. While all the products from one vendor can be managed through say, an SNMP (simple network management protocol) system, you may not have the ability from a central location to tap all the features in each system's proprietary management system.
Capacity a Key
There are nearly 40 vendors currently selling Ethernet switching hubs. But fewer than a dozen are poised to enter the ATM-to-Ethernet switching market, because of the complexity of merging the two technologies. In an ATM network, the Ethernet packets or frames must be converted to ATM cells (and vice versa). This packet-to-cell conversion technology is not widely available today. The conversion is done using an ASIC (application specific integrated circuit), which only a handful of Ethernet switch vendors have developed--notably, Nicecom (Lexington, MA), which was acquired by 3Com, and Onet.
Another key factor to consider is how the Ethernet switch operates under great loads. Depending on the networking environment into which the switch is placed, this can either be a major problem or no problem at all. For example, it's likely that companies taking advantage of merging ATM and Ethernet switching will connect multiple Ethernet LAN segments to high-performance servers. This setup, in which servers are put onto the high-speed backbone and the clients remain on Ethernet LANs, is commonly called a server farm.
While this seems like an ideal situation, you'll need to be concerned about blocking, which occurs when two clients on different LAN segments contend for the same server at the same time. If a client on one LAN segment is passing a large amount of data to a server and a second client tries to send data, a buffer in the Ethernet switch will hold the data from the second client until the server can accept it.
If a third client tries to have an exchange with another server while the second client's data is held in the buffer, most of the Ethernet switching hubs on the market will hold the third client's data in a queue behind the already buffered data. Some vendors, such as Onet, have circumvented this problem by developing nonblocking hubs that let the data destined for another server jump ahead of the buffered data in the queue, so that it may be passed to the free server.
Since most Ethernet switching hub vendors do not have ATM interfaces developed, it is not possible today to tell whether their products will suffer from blocking or not. However, users will soon be able to sort these issues out as more products make it to market. The hardware is becoming available and some of the management software is here. That bodes well for the merging of the two technologies, and it should help network managers provide the additional bandwidth they'll need for new applications without having to do a forklift overhaul to their networks.
ISSUE CONSIDERATIONS Network architecture Must choose between a modular-based enterprise hub or a stand-alone workgroup hub with interface to an ATM switch Integrated management system Make sure management system offers full feature set for all devices (many vendors have acquired switching products to round out their product line and have not integrated the management systems) ATM interface Make sure stand-alone hub vendor has solid technical relationship with one or two of the major ATM switch vendors Conversion Look for vendors with ASIC technology that performs ATM cell to Ethernet frame conversion Capacity under stress Look for distributed processor architecture that is nonblocking
One way to economically increase network bandwidth is to combine ATM (asynchronous transfer mode) switches with Ethernet switching hubs.