Your best bet in faster LANs A walk though Networld+Interop reveals a faster network need not cost a bundle By Robert E. Lee

October  1996

Mail this article to a friend

Atlanta -- Sluggish networks dragging you down? Maybe the new solutions decked out at Network+Interop in mid-September didn't get hearts racing, but they did at least raise some eyebrows. Lo and behold it seems ATM has finally entered the mainstream as a technology to span the desktop and the enterprise across the WAN. Fast Ethernet and 100VG have become rather common place and are being widely deployed, while switching technologies have upped the throughput of every network speed. Then there's Gigabit Ethernet, which offers 1000-megabit-per-second networking speed on an Ethernet format.

All of these technologies were spotlighted at the recent Networld+Interop tradeshow. We'll get to that later. Your real challenge now is the development of a logical choice of networking alternatives to move you from the 10-megabit-per-second congested network you have today to the ultra-fast speeds each of the technologies above offer. First, let's take a step back and review the topology of a network and the implications of various design choices.

Originally Ethernet networks came on coaxial cable, in both a thick and thin version. The transport technology in these cables effected the distance and packet collision characteristics of the network. But one major issue emerged in this arrangement -- the network was severely hampered by the use of this single cable, since the network had to be wired from one device to another in serial fashion. If a network device was moved, the network had to be separated from the device, and then the cable had to be spliced with a connector to replace that device. When a new device was added, the cable either had to be split or a new cable had to be pulled to recreate this linear topology.

Enter 10BaseT. This technology began as a way to reuse telephone cable and to wire facilities with data and voice in a single wire pair. The topology often resembled a star, where a single hub would have many cables emanating out from the hub to each device. The hub would in turn be connected to another hub, which might be connected to the server and other network devices. This design allowed for the addition and deletion of devices without disrupting the network or forcing you to recable while the network was down.

In both of these technology implementations, the network packets travel across the entire length of the network, seen by all devices as they listen to find the traffic intended for themselves. This process led to the concept of collisions, where more packets existed on the wire than the devices could retrieve at the speed of the network. Solutions were needed, and Ethernet-switch technology was born.

On to Ethernet-switching
Ethernet-switching is a concept that extended the speed of the standard 10-megabit-per-second backbone by recognizing two factors. Backplanes in the switches could run faster than 10 megabits per second, and not all packets were destined for all devices. What the switch does, in a simple view, is to dynamically route each packet between ports on the switch, sending the packet down only the wire(s) that have the destination device identified in the packet. This technique has provided the network the illusion that each device has a dedicated 10-megabit-per-second network, even though that is not fully true. If a network is peer-to-peer, this illusion is a near reality, but in a strong client/server model, the server becomes saturated with all of the 10-megabit-per-second feeds from the clients, limiting the network to 10-megabit-per-second overall throughput.

Ethernet-switches can be implemented without regard to the network interface card on the computer. Since these switches support all popular protocols, speeding up the network through this technology is a fast fix and a logical first step in the evolution to higher network speeds.

But simple switching technologies are not enough. As you saw in the previous model, client/server networks didn't benefit much from this technology. Now is the time to step up to 100-megabit-per-second networking. Two approaches exist in the market today, Fast Ethernet (an implementation of existing Ethernet technology at 10 times the speed) and 100VG (a priority demand derivative of token ring). Both claim extensive throughput approaching 90 megabits or better with reduced collision and improved throughput.

Mixing Ethernet-switching with 100-megabit-per-second Ethernet can yield high-performance networks in both peer-to-peer and client/server models. Servers and high-performance workstations can be placed on the network at 100 megabits per second while other workstations can be left at 10 megabits per second through an Ethernet-switch, increasing local throughput to a full 10 megabits per second. Sun has standardized on the 100-megabit-per-second Fast Ethernet for on-board network interfaces and as an add-on to older workstations and servers.

But 100 megabits per second still does not satisfy the demand for bandwidth. As the number of devices entering the network at 100 megabits per second increases, the demand for an even faster backbone in the network rears its ugly head. Think of the following analogy, a series of streams entering a river. If each stream has an upper limit of 100 gallons per second, and there are 50 of those streams entering the main channel, than the maximum flow into the main channel would be 5000 gallons per second. If the main channel is limited to the same 100 gallons per second, then all feeding channels will be limited to 1/50 of their rated speeds.

It is not a reasonable expectation that each stream will flow at full capacity at all times. Just as you view situations like floods at points of every 1-, 10-, and 100-year levels, so must the network be configured. You must plan for normal needs. In configuring your backbone, recognize that each link to the LAN or WAN will operate at maximum speed for a fraction of the time and that the chance of all links requiring full bandwidth simultaneously is small.

The need for speed
In order to boost the speed of the backbone, it is necessary to employ one of two technologies, ATM or Gigabit Ethernet. Asynchronous Transfer Mode (ATM) provides a 53-byte cell that travels across the physical layer of the network, scalable to speeds of 622 megabits per second and eventually beyond. Current network cards provide for 622-megabit ATM for Sun workstations, providing incredible bandwidth for applications like full-motion video, imaging, and computational/database-intensive operations across systems. Gigabit Ethernet provides for 1000 megabits per second of data rate in an Ethernet format. This technology has solved many of the problems in 10/100-megabit-per-second Ethernet, opening the backbone to this higher level.

The question becomes one of which technology is best for your backbone needs. The answer is of course, both, based on your network goals. Gigabit Ethernet is an excellent choice for campus networks where high speeds are required for data applications and the gateways to the WAN do require the same levels of throughput. The Gigabit Ethernet Alliance reports that expected prices per port for this technology will be in the $920 to $1400 range by 1998.

ATM at 622 megabits per second on the other hand is expected to cost about $4,200 per port in 1998. This discrepancy in cost stems from the greater flexibility in ATM to carry voice, video, and data traffic. Across the WAN, ATM can provide integration of all of your voice circuits with data, providing a packet stream that reduces overall communications costs by sharing bandwidth on the network. A new round of ATM technologies from Sun, Cisco, Bay Networks, and others brings this technology closer to the returned promise of voice and data sharing the network.

Let's review your basic strategies:

• Start with 10-megabit-per-second Ethernet-switch technology to get short-term performance gains with existing hardware.

• Move to 100-megabit-per-second transport technologies for the backbone and high performance workstations and servers combined with 10-megabit-per-second Ethernet-switch hubs.

• Once 100 megabits per second has become the predominate transport speed, move the backbone and transports to ATM if wide-area connectivity is the dominate network link. Use Gigabit Ethernet if the traffic is on the LAN in a campus environment.

One more note, while your category 5 cabling solutions can provide the physical transport for the ATM and Gigabit technology, the distance limitations are severe -- 100 meters or less. Fiber optics becomes the cable of choice in these situations, so plan accordingly.

Networld+Interop goods
Here's a brief recap of the relative cost per port of each technology demonstrated at the Networld+Interop show for a sample of vendors. This overview includes a basic model and then the fully populated switch cost with the maximum number of ports installed. The cost per port increases dramatically as you reduce the number of ports installed because the switch carries a baseline cost. For example, the Cisco LightStream 1010 is $19,000 for the switch, without any ports. If you were to install just four ports of the ATM 155 cards, the per port price is $5,275 while the fully populated switch is only $1,120. The range of port prices comes from the choices for physical media, typically Unshielded Twisted Pair Category 5, Single, or Multimode Fiber.

Gigabit Ethernet is a technology that exists in the backplane of today's switches and will move out to the port side early next year. The standards are still being development and chipsets finalized. The Gigabit Ethernet Alliance has the scoop on where this technology is at and who will be supporting it, including major network providers like Hewlett-Packard and Cisco.

As you can see, additional speed requirements escalate prices dramatically, but relationship between price and performance is relatively constant or improves as you move to each higher level. Each company offers a wide range of product options to scale from simple workgroups to enterprise-wide solutions. For a more comprehensive set of resources on these networking alternatives, drop by each company's Web site and in particular the Networld+Interop site.
--Robert E. Lee

Resources

• "(Barely) managing ATM"
  /sunworldonline/swol-08-1996/swol-08-atm.html
• "Stretching the network"
  /sunworldonline/swol-08-1995/swol-08-ethernet.html
• "Have VLANs progressed?"
  /sunworldonline/swol-07-1996/swol-07-vlan.html
• ATM at Sun
  http://www.sun.com/ATM/
• Cisco Systems
  http://www.cisco.com/
• Networld+Interop
  http://www.interop.com/
• Bay Networks
  http://www.baynetworks.com/
• Fore Systems
  http://www.fore.com/
• Hewlett-Packard
  http://www.hp.com/
• ATM Forum
  http://www.atmforum.com/
• Gigabit Ethernet Alliance
  http://www.gigabit-ethernet.org/

If you have technical problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-10-1996/swol-10-networking.html
Last modified: