MORE RISING STARS

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

Yes, it's time for another exciting episode of the Trade Show Tango. My desk is overflowing with imaginative new products and major revisions to old ones, so I decided to devote another column to the rising stars of the bunch.

Before I mention the first product, a technical explanation is needed. Switched Ethernet is a technique of increasing the amount of data that can be handled by an Ethernet network. Ethernet is essentially one big party line. (Remember the hand-cranked phones in the original Lassie Series?)

When one computer is about to send a message, it first listens to the party line to hear if anyone else is already sending one. Assuming the line is free, it tries to send its message. When it talks, all of the other computers hear it, not just the intended receiver. If more than one computer tries to talk at the same time, the other computers notice the increased-strength signal caused by the crisscrossing original signals. The first one to notice this signal increase puts out a special jam signal to tell everybody to back off and try again. (Think of the buzzer that sounds off when someone answers a question incorrectly on a game show.)

When the other computers hear the jam signal, any of them that were in the middle of trying to send a message stop immediately, wait for while, and then try to send again. The definition of "a while" is determined by a random number generator built into the Ethernet card.

For those who aren't fans of programming, a random number generator gives a different result number for each "seed" number that it is fed. For this process, the seed is the Ethernet identification number of each network card. Barring a major manufacturing error (at least one of these errors has occurred), every Ethernet card ever made has a different Ethernet ID. A different ID means a different seed number and, therefore, a different result number. Since the result number determines the definition of the "while" that each card waits before resending its message, each card will wait a different amount of time before resending, thereby avoiding another simultaneous send.

This process is complicated by the fact that each computer's ability to resend is affected by its own speed and by the relative efficiency of its Ethernet card, but in practice it works fairly close to the theoretical model. The overall process is called CSMA/CD, meaning Carrier Sense Multiple Access with Collision Detection. (Love those buzzwords!)

Ethernet provides a theoretical maximum of ten million bits per second (where eight bits make a byte which is the equivalent to a single character), but this is shared via the aforementioned process by all of the computers on the network. Switched Ethernet is a sneaky way around this sharing problem.

Switched Ethernet uses the same basic principles used by the phone company before fully automated switching was available. Upon request, an operator would have to manually place a jumper cable between the sockets leading to the sending and receiving phones, creating a temporary complete circuit. (Think of Lily Tomlin's Ernestine character.) After the call was complete, the jumper would be removed.

In the modern version of this example, a temporary circuit is created between the sending and receiving computer so fast that the two essentially have the entire ten million bits per second to themselves for just long enough to get the message through. Then the line is freed up for the next pair of computers to communicate.

If you make the assumption that this particular operator in our example is handling the connection point between a single trunk line and multiple local lines, the comparison with switched Ethernet on a LAN is more accurate. (Though I am sure that techies reading this are already squirming in their seats over this gross oversimplification.) Since most of the traffic on a LAN is between the workstations and the file server, the line to the file server becomes the "trunk line" in this example. It is the line through which most data must pass and is the one that we must optimize the most.

This leads me (finally) to the first of the hot new products, a clever idea developed by XNET Technology (Milpitas, CA, 408-263-6888). They have developed a product that they call a ParallelSwitch, which encapsulates an Ethernet switching hub on a single ISA (Industry Standard Architecture) or EISA (Extended ISA) board. It contains one basic Ethernet chipset to talk to the file server's motherboard (the trunk line) and six switched Ethernet ports to service the network (the local lines).

While switching is not an original concept, the idea of putting what amounts to a complete switching hub on a single board is. At $1,988 for ISA and $2,988 for EISA 10Base-T models, this product is a major price breakthrough in this category. In addition, it is arguably more efficient than an external switching hub since the switching circuitry is on the same board as the servers own Ethernet chipset.

Each time the file server communicates (in either direction) over one of the six switched lines, it momentarily sees the full bandwidth (essentially, how much room for signals) of Ethernet. This raises the theoretical effective throughput (how many signals really gets through) to sixty million bits per second!

Of course the real world will intrude and keep us from getting the theoretical ideal, but we can maximize our throughput by selecting EISA as opposed to ISA. With an EISA interface, we are less likely to run into bottlenecks in throughput due to limitations within the underlying bus architecture. Within those limitations, XNET's solution to a maxed-out Ethernet is very useful, especially for the network administrator with a limited budget.

The second product on my list is really two similar products, 3Pack from Funk Software, Inc. (Cambridge, MA, 800-828-4146), and NetSqueeze from the LAN Support Group (Houston, TX, 800-749-8439). Both are file compression NLMs (Netware Loadable Module) for Netware 3.11 and 3.12 file servers. Both mimic the configurable compression features built into Netware 4.x, even to the point of using licensed STAC compression.

These products provide a way for an administrator to take advantage of compression without making the transition to Netware 4. While I suspect that both companies may be getting hate-mail from Novell for encouraging users to delay upgrading, I have to admit that I regularly find companies that are not quite ready - either technologically or financially - to upgrade. Aftermarket compression products are a method of letting an old system limp along until the inevitable (in my opinion) upgrade becomes feasible.

The last two products on my list are actually upgrades to ones that I have previously praised. Micropolis Corporation (Chatsworth CA, 800-395-3748) has made major improvements in its Radion series of RAID (Redundant Array of Inexpensive Disks) systems. They have designed a controller that they call Gandiva, to give hardware RAID to their systems. You can even start out with software-based RAID and move up to hardware-based later. This improvement will make their Radion Series much more competitive with the high-end RAID systems currently available, as well as providing flexibility due its upgrade capability.

Another enhancement that Micropolis introduced is a rack-mounted version of its RAID system. This RAID will allow expansion to 81GB using 3.5" drives or an amazing 245GB using 5.25" drives! The system should be handled by the Gandiva controller to enhance speed (though technically you could use software). This should keep staunch mainframe advocates up worrying all night!

The final product is the new CommSwitch Series 2000 scalable computer system from CommVision (Mountain View, CA, 415-254-5270). I have written about CommVision's (formerly IWI) eight and twenty slot computer systems before. Now they have a rack-mountable very expandable version available. They have versions with up to 180 32-bit, hot-swappable slots! Up to 160 of these slots can be filled with processor boards. Talk about multiprocessing!

Since Netware is already divided into segmentable pieces called NLMs, it is quite easy to split Netware's load (using Run-Time Netware) among many processors without the efficiency bottlenecks and reprogramming necessary to fully implement symmetrical multiprocessing systems. Each of these processors can be a 486 or a Pentium with up to 128MB RAM for the 486s and 256MB RAM for the Pentiums. Each board is hot-swappable and has a full 32-bit, high-speed interface. The overall packet-switching bus that connects these boards has an aggregate peak transfer rate of up to 8.5Gb (that's Gigabits) per second!

While CommVision is billing this system as the ultimate communications server, I would suggest combining it with hardware RAID (maybe from Micropolis) and turning it into an incredibly fault tolerant superserver. I suspect that it would run circles around conventional superservers with only four or so processors, and throw in the added reliability as a bonus.

Finally, I just wanted to make it clear that there is no truth to the rumor that the Justice Department has been purchased outright by Bill Gates in return for settling the national debt.

1995, Wayne M. Krakau