SPEEDY ALKA-SELTZER

by Wayne M. Krakau - Chicago Computer Guide, May 1995

No, this is not a nostalgic tribute to classic advertising campaigns. (Besides, if I remember correctly, I was still wearing pajamas with feet when that campaign was in its prime.) The "speedy" in the title refers to LAN speed at the physical level - just how fast can you pump data down the pipe. The "Alka-Seltzer" refers to the medication that LAN administrators (and systems integrators!) are going to need when they find out that their already complicated lives just got a lot more complex.

FDDI (Fiber Distributed Data Interface), a 100Mbps (millions of bits per second) descendant of Token Ring, has been available for several years, but is still too pricey at $1300 and up per card to make much of a dent in the workstation connection market. It has been used for high-speed backbones, but is still a bit expensive even for that.

ATM (Asynchronous Transfer Mode, not Automated Teller Machines) provides a 155Mbps link, but it is too new to be in widespread use and makes FDDI look inexpensive. IBM’s special 25Mbps version, called ATM25, is seen as being too immature, even though it is in the process of being approved as a secondary standard. While it is less expensive than the 155Mbps version of ATM, it suffers in comparison with other methods in a cost per megabit comparison.

Administrators of Ethernet LANs are being swamped with new options to wring more speed out of a network connection. The real challenge is to decide amongst them.

The most spectacular methods are two that involve raw speed, 100Mbps versus the 10Mbps Ethernet standard. These two methods are called 100Base-T and 100VG-AnyLAN, and are designed to work, initially, over unshielded twisted pair (UTP) copper wire. If you know that Ethernet, when using UTP, is called 10Base-T, you can get an idea of these names’ derivations.

Each of the two competing standards (or should I say "almost" standards, since as I write this, neither is technically finalized) is supported by its own consortium of manufacturers, sixty-two for 100Base-T and twenty-six for 100VG-AnyLAN, with five companies straddling the fence by supporting both types (by my count in mid April 1995). There was supposed to be only one standard called "Fast Ethernet", but the political maneuvering got so messy that we ended up with two.

100Base-T is the true descendant of 10Base-T, so it inherited the moniker "Fast Ethernet". It follows the same basic methods as 10Base-T, in particular CSMA-CD (Carrier Sense Multiple Access with Collision Detection), its way of arbitrating which computer gets access to the LAN. The main difference is that 100Base-T’s raw speed is ten times as high and the maximum span of the LAN is correspondingly only one-tenth of its predecessor’s. That gives you between 200 and 210 meters to work with, depending on the type of concentrator or hub that you use. You are also restricted to only two repeaters or hubs.

Currently, Fast Ethernet works only with Level 5 (the highest quality) cable using two pairs of a four-pair bundle. A standard (Yikes, more standards!) for using the older, more common Level 3 UTP cable by utilizing all four pairs of a four-pair bundle is on the way. Concurrently, the 100Base-FX standard for fiberoptic cable is being finalized. Using fiber segments will help alleviate the problems caused by the severe distance limitations of Fast Ethernet.

As they say on Monty Python, "And now for something completely different." The other standard, 100VG-AnyLAN, really doesn’t have a lot to do with the original Ethernet standard. It is a new hybrid design with bits of Ethernet, Token Ring, and various other methodologies. It maintains the typical Ethernet distance limitations and can work with Level 5 (two pairs) or Level 3 (either two or four pairs) UTP cable, and soon, with fiberoptic cable. Because VG eliminates potential collisions within its hubs, it tends to do better in applications where a continuous stream of data is needed, such as video or sound transmission. Its "Ethernet" name is basically a marketing ploy, implemented to make this new standard more palatable to LAN managers and upper management types. (Note that the preceding sentence is a disputable political statement of my opinion. It’s considered the "purist" point of view.)

An older technology that is just hitting its stride is Ethernet switching technology. For a detailed explanation of this technology see my More Rising Stars column from two months ago. In summary, the hub gives the full Ethernet 10Mbps of bandwidth to one port at a time by switching between ports at very high speeds. At the other end of those ports, you can put other concentrators, or for maximum throughput, wire them directly to individual workstations. Keep in mind, however, that Ethernet switching is totally proprietary - each vendor has their own unique way of doing it.

Yet another technology (Hey, I told you this would make your life more complex!) is Duplex Ethernet. It allows simultaneous two-way communications over the network cable. This concept is similar to the comparison between half-duplex and full-duplex within your communications software. This is advantageous when there is a lot of data being sent by workstations. Since most networks have 95% or more of the data coming from the file server, this isn’t necessarily a great advantage. It is advertised as providing a theoretical 20Mbps channel, but your mileage may differ. Just to keep things interesting, Duplex 100Base-T is on the way. It raises the ante to a theoretical 200Mbps. Luckily for those evaluating this technology, Duplex Ethernet is interoperable among different brands of hubs and network cards.

Finally, (Now that I’ve got your head really spinning.) multiport Ethernet cards for file servers are going to get very popular due to the potential speed of the PCI bus and the burgeoning 100Base-T market. Manufacturers are developing four-port PCI Duplex Fast Ethernet cards with a theoretical aggregate throughput of 800Mbps! Each of these cards takes up only one slot in the file server, so you can use more than one. Network card design has advanced to the point where several of these cards placed in one file server are efficient enough that they won’t bog down the filer server’s processor.

Once all of the hardware is available, a maximized system could combine these technologies, but the best use of these technologies is probably in some combination based on the availability of new products and tempered by real world cost considerations. For example, I am currently designing a system in which a compromise solution has the best price-performance ratio. It will include two four-port 100Base-T PCI cards leading to separate backbone lines to each wiring closet. Within each wiring closet there will be an Ethernet switching hub with a single 100Base-T port on its back and multiple 10Base-T ports on its front. The workstations will use standard 10Base-T cards. One great advantage of this design is its expandability. With the 100Mbps backbone lines, we can upgrade to 100Base-T concentrators if they’re ever needed in the future.

Now that I’ve finished covering these new toys, I’m going to the drug store to stock up on more of my favorite analgesics and antacids.

1995, Wayne M. Krakau